Method for detecting and analyzing tumor angiogenesis based on multimodal image fusion

Abstract

The present application discloses a method for detecting and analyzing tumor angiogenesis based on multimodal image fusion, which relates to the field of medical imaging technology and includes: collecting multimodal image data of a target tumor; extracting features from the collected multimodal image data according to a neural network model; fusing the extracted features through a fusion network to obtain a comprehensive feature vector; analyzing the tumor angiogenesis status according to the comprehensive feature vector to obtain an analysis result; and evaluating the angiogenesis status of the tumor and generating a report according to the analysis result. It can provide more comprehensive tumor angiogenesis information and improve the accuracy of detection and analysis. Through the fusion of multimodal image data, it can more realistically simulate the in-vivo tumor microenvironment. It can be used as a preclinical research platform for anti-tumor and anti-tumor angiogenesis drugs to provide support for drug screening and evaluation.

Description

Technical Field

[0001] This application relates to the field of medical imaging technology, and particularly to a method for detecting and analyzing tumor angiogenesis based on multimodal image fusion. Background Art

[0002] In the research of tumor angiogenesis, the traditional two-dimensional in vitro vascular model is very different from the in vivo tumor microenvironment and cannot simulate the three-dimensional structure of in vivo tumors, resulting in great obstacles to the research on tumor angiogenesis and normalization. In addition, the existing 3D tumor models and tumor organoids do not meet the conditions for new blood vessels, restricting the in-depth study of the mechanism of tumor angiogenesis.

[0003] With the development of modern medicine, medical images (such as X-ray films, CT scans, MRI, etc.) have become important tools for doctors to diagnose and treat diseases. Doctors need to accurately analyze and understand medical images to formulate correct diagnostic and treatment plans. However, the analysis and interpretation of medical images require a lot of time and energy from doctors. Especially when faced with a large number of images, it is difficult for doctors to maintain high-quality analysis and interpretation. In addition, different doctors have different experience and skill levels, which may also lead to different interpretations and diagnoses of the same image.

[0004] How to improve the auxiliary level of the diagnostic prediction system through medical images has become an urgent problem to be solved. Summary of the Invention

[0005] Therefore, the present invention aims to provide a method for detecting and analyzing tumor angiogenesis based on multimodal image fusion. This method can provide more comprehensive tumor angiogenesis information by fusing data from different imaging modalities (such as CT, MRI, PET, etc.), and improve the detection and analysis ability of tumor angiogenesis through algorithm optimization.

[0006] To achieve the above object, the present application provides the following technical solutions:

[0007] A method for detecting and analyzing tumor angiogenesis based on multimodal image fusion, comprising:

[0008] S1. Collect multimodal image data of the target tumor;

[0009] S2. Extract features from the collected multimodal image data according to a neural network model;

[0010] S3. Fuse the extracted features through a fusion network to obtain a comprehensive feature vector;

[0011] S4. Analyze the tumor angiogenesis state according to the comprehensive feature vector to obtain an analysis result;

[0012] S5. Evolve the three-dimensional tumor blood vessel model according to the comprehensive feature vector, and fuse the analysis results with the three-dimensional tumor blood vessel model;

[0013] S6. Evaluate the angiogenesis status of the tumor based on the analysis results and the fused three-dimensional model, and generate a report.

[0014] Further, S1 further includes segmenting the multi-modal image data of the target tumor based on a conditional loss function, specifically: segmenting the region of the target tumor based on the conditional loss function;

[0015] The loss function model is:

[0016] ;

[0017] where l ct represents the real tumor image, m ct represents the real tumor segmentation label, l p represents the pseudo-liver image generated by the generator, m p and m ct represent the tumor segmentation label predicted based on the blood vessel structure and the real tumor segmentation label respectively, m pi and m cti represent the i-th tumor segmentation label predicted based on the blood vessel structure and the real tumor segmentation label respectively; represents the absolute value of the pixel-by-pixel difference between the real tumor image and the pseudo-tumor image, H, W, and D represent the height, width, and depth of the tumor image respectively, represents the pixel value of the real tumor image, represents the pixel value of the pseudo-tumor image, represents the binary cross-entropy loss, represents the DSC loss, Ls represents the image loss, and N is the number of images for data acquisition, 、 , represent the weight values respectively.

[0018] Further, feature extraction is performed on the segmented multi-modal image data according to the neural network model, including: the tumor image matching and deformation process and the neural network model feature extraction process;

[0019] Further, the tumor image matching and deformation process includes:

[0020] S21. First, extract key points from the input image. The key points include corner points, edges, and textures;

[0021] S22. After the key points, match the key points in the original image and the target image;

[0022] S23. After the key point matching is completed, calculate the deformation matrix between the original image and the target image; the deformation matrix characterizes the distance and direction that each pixel point needs to move to align the original image with the target image.

[0023] S24. Match and deform the source image according to the calculated deformation matrix.

[0024] Wherein, the original image is an MRI tumor image or a PET tumor image, and the target image is a CT tumor image.

[0025] Further, the feature extraction process of the neural network model includes:

[0026] Use a convolutional neural network with the AlexNet structure to extract the image features of the MRI tumor image, the PET tumor image, and the CT tumor image. Set the 3D convolutional kernel to move sequentially according to a preset step size, and perform a dot product operation on the pixel values of the image at the corresponding covered positions. The result corresponds to a pixel point in the output image, and multiple pixel points after the operation form a feature map.

[0027] Use a pooling layer to perform average pooling on the feature map to obtain a target feature map, and obtain a feature vector through the target feature map, where the feature vector is obtained from the feature map through a flattening operation.

[0028] Further, fuse the extracted features through a fusion network. Among them, the fusion network includes a first-layer network and a second-layer network. The first-layer network fuses the features of different modalities from the micro to the macro, that is, the first-layer network fuses the features of different modalities from high resolution to low resolution. The second-layer network fuses the pairwise-fused features to obtain a first comprehensive feature and a second comprehensive feature. The steps to obtain the first comprehensive feature include:

[0029] In the first-layer network, splice the feature vectors of the MRI tumor image and the PET tumor image to obtain a first comprehensive feature A.

[0030] For example, if the MRI feature vector is [F MRI1 , F MRI2 ,..., F MRIn , and the PET feature vector is [F PET1 , F PET2 ,..., F PETm , the spliced feature vector is [F MRI1 , F MRI2 ,..., F MRIn , F PET1 , F PET2 ,..., F PETm .

[0031] Optionally, in the first layer of the network, the feature vectors of the MRI tumor image and the CT tumor image are concatenated, that is, the first comprehensive feature B,

[0032] Optionally, in the first layer of the network, the feature vectors of the PET tumor image and the CT tumor image are concatenated to obtain a first comprehensive feature C.

[0033] S32, the comprehensive feature vector output by the first layer network is sent to the second layer network to obtain the second comprehensive feature; the second layer network includes a fully connected layer and several convolutional layers to integrate and abstract features;

[0034] Optionally, the first comprehensive features A, B, and C are weighted averaged and input into the second layer network to obtain a second comprehensive feature;

[0035] S33, the second comprehensive feature is used in the softmax layer to generate a predicted blood vessel segmentation result;

[0036] Optionally, the blood vessel segmentation results obtained by the second comprehensive features generated by the first comprehensive features A, B, and C are weighted averaged to make the blood vessel model smoother;

[0037] Furthermore, the tumor angiogenesis state is analyzed according to the comprehensive feature vector to obtain analysis results, which include: vascular density, vascular structure, evaluation of vascular function, vascular inner diameter and carotid intima-media thickness, and location of atherosclerotic plaques.

[0038] Furthermore, the tumor vascular three-dimensional model is evolved according to the comprehensive feature vector, and the analysis result is integrated with the tumor vascular three-dimensional model, including:

[0039] The blood vessel segmentation result is obtained according to the comprehensive feature vector, and a three-dimensional blood vessel model of the tumor area is generated according to the blood vessel segmentation result. The analysis result is color mapped so that the three-dimensional blood vessel model of the tumor area presents different colors.

[0040] Further, evaluating the angiogenesis state of the tumor includes randomly selecting the coordinate values and pixel values of several feature points of the three-dimensional model of the blood vessels in the tumor area and comparing them with the coordinate values and pixel values of the corresponding points on the real scanned image; calculating the cumulative value of the pixel value deviation value and the cumulative value of the coordinate value deviation value; the feature point is the arc protrusion point or the end position point of the blood vessel,

[0041] Furthermore, the evaluation value is: a weighted value of the accumulated value of the pixel value deviation value and the accumulated value of the coordinate value deviation value.

[0042] Furthermore, when the accumulated value of the pixel value deviation, the accumulated value of the coordinate value deviation, the accumulated value of the pixel value deviation and the accumulated value of the coordinate value deviation are all less than the threshold value, the model is qualified.

[0043] A tumor angiogenesis detection and analysis system based on multimodal image fusion, which is used to execute the tumor angiogenesis detection and analysis method based on multimodal image fusion described in any one of the above, includes:

[0044] An image acquisition module, which is used to acquire multimodal image data of a target tumor;

[0045] A feature extraction module, which is used to extract features from the acquired multimodal image data according to a neural network model;

[0046] An image fusion module, which is used to fuse the extracted features through a fusion network to obtain a comprehensive feature vector;

[0047] An image analysis module, which is used to analyze the tumor angiogenesis state according to the comprehensive feature vector to obtain an analysis result;

[0048] A model generation module, which is used to evolve a three-dimensional model of tumor blood vessels according to the comprehensive feature vector and fuse the analysis result with the three-dimensional model of tumor blood vessels;

[0049] A report generation module, which is used to evaluate the tumor angiogenesis state according to the analysis result and the fused three-dimensional model, and generate a report.

[0050] Beneficial effects

[0051] It can provide more comprehensive tumor angiogenesis information and improve the accuracy of detection and analysis. Through the fusion of multimodal image data, it can more realistically simulate the in-vivo tumor microenvironment. It can be used as a preclinical research platform for anti-tumor and anti-tumor angiogenesis drugs, providing support for drug screening and evaluation. Description of the drawings

[0052] To more intuitively illustrate the prior art and this application, several exemplary drawings are given below. It should be understood that the specific shapes and structures shown in the drawings generally should not be regarded as limiting conditions when implementing this application; for example, those skilled in the art are capable of making routine adjustments or further optimizations to the addition / deletion / attribution division of certain units (components), specific shapes, positional relationships, connection methods, dimensional proportional relationships, etc. based on the technical concept disclosed in this application and the exemplary drawings.

[0053] Figure 1 It is a flowchart of the tumor angiogenesis detection and analysis method based on multimodal image fusion.

[0054] Figure 2 It is a structural diagram of the tumor angiogenesis detection and analysis system based on multimodal image fusion. Detailed implementation manners

[0055] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0056] In the description of the present application: Unless otherwise specified, "a plurality of" means two or more. Terms such as "first", "second", "third", etc. in the present application are intended to distinguish the objects being referred to, and do not have special meanings in terms of technical connotations (for example, they should not be understood as emphasizing importance or order, etc.). Expressions such as "including", "comprising", "having", etc. also mean "not limited to" (certain units, components, materials, steps, etc.).

[0057] Terms such as "upper", "lower", "left", "right", "middle", etc. cited in the present application are usually for the convenience of visually understanding with reference to the accompanying drawings, and are not absolute limitations on the positional relationships in actual products. Without departing from the technical concept disclosed in the present application, changes in these relative positional relationships should also be regarded as falling within the scope of the expression of the present application.

[0058] Embodiment

[0059] A method for detecting and analyzing tumor angiogenesis based on multimodal image fusion includes:

[0060] S1. Collect multimodal image data of the target tumor;

[0061] S2. Extract features from the collected multimodal image data according to the neural network model;

[0062] S3. Fuse the extracted features through a fusion network to obtain a comprehensive feature vector;

[0063] S4. Analyze the tumor angiogenesis status according to the comprehensive feature vector to obtain an analysis result;

[0064] S5. Evolve the three-dimensional model of the tumor blood vessels according to the comprehensive feature vector, and fuse the analysis result with the three-dimensional model of the tumor blood vessels;

[0065] S6. Evaluate the tumor angiogenesis status according to the analysis result and the fused three-dimensional model, and generate a report.

[0066] S1 further includes segmenting the collected multimodal image data of the target tumor based on a conditional loss function, specifically: segmenting the region of the target tumor based on the conditional loss function;

[0067] The loss function model is:

[0068] ;

[0069] where l ct represents the real tumor image, m ctRepresents the true tumor segmentation label, l p Represents the pseudo-liver image generated by the generator, m p And m ct Indicates that they respectively represent the tumor segmentation label predicted based on the vascular structure and the true tumor segmentation label, m pi And m cti Respectively represent the i-th tumor segmentation label predicted based on the vascular structure and the true tumor segmentation label; Indicates the absolute value of the per-pixel difference between the true tumor image and the pseudo-tumor image. H, W, and D respectively represent the height, width, and depth of the tumor image, Represents the pixel value of the true tumor image, Represents the pixel value of the pseudo-tumor image, Represents the binary cross-entropy loss, Represents the DSC loss, Ls represents the image loss, and N is the number of images for data acquisition, 、 , Respectively represent the weight values.

[0070] Furthermore, feature extraction is performed on the segmented multi-modal image data according to the neural network model, including: the tumor image matching deformation process and the neural network model feature extraction process;

[0071] Furthermore, the tumor image matching deformation process includes:

[0072] S21. First, key points are extracted from the input image. The key points include corner points, edges, and textures;

[0073] S22. After extracting the key points, the key points in the original image and the target image are matched;

[0074] S23. After the key point matching is completed, the deformation matrix between the original image and the target image is calculated; The deformation matrix characterizes the distance and direction that each pixel point needs to move to align the original image with the target image;

[0075] S24. The source image is matched and deformed according to the calculated deformation matrix;

[0076] Where the original image is the MRI tumor image, the PET tumor image, and the target image is the CT tumor image;

[0077] Furthermore, the neural network model feature extraction process includes:

[0078] The convolutional neural network adopting the AlexNet structure extracts the image features of MRI tumor images, PET tumor images, and CT tumor images. Set the 3D convolutional kernel to move sequentially at a preset stride, perform a dot product operation on the pixel values of the images at the corresponding covered positions, and the result corresponds to a pixel point in the output image. After the operation, multiple pixel points form a feature map;

[0079] Adopt a pooling layer to perform average pooling on the feature map to obtain the target feature map, and obtain the feature vector through the target feature map, where the feature vector is obtained from the feature map through a flattening operation.

[0080] Furthermore, fuse the extracted features through a fusion network. The fusion network includes a first-layer network and a second-layer network. The first-layer network fuses the features of different modalities pairwise from the micro to the macro; the first-layer network fuses the features of different modalities pairwise from high resolution to low resolution, and the second-layer network fuses the pairwise fused features to obtain the first comprehensive feature and the second comprehensive feature; the first comprehensive feature obtained includes:

[0081] S31. In the first-layer network, splice the feature vectors of the MRI tumor image and the PET tumor image to obtain the first comprehensive feature A,

[0082] If the MRI feature vector is [F MRI1 , F MRI2 ,..., F MRIn , and the PET feature vector is [F PET1 , F PET2 ,..., F PETm , the spliced feature vector is [F MRI1 , F MRI2 ,..., F MRIn , F PET1 , F PET2 ,..., F PETm ;

[0083] Optionally, in the first-layer network, splice the feature vectors of the MRI tumor image and the CT tumor image, that is, the first comprehensive feature B,

[0084] Optionally, in the first-layer network, splice the feature vectors of the PET tumor image and the CT tumor image to obtain the first comprehensive feature C,

[0085] S32. The comprehensive feature vector output by the first-layer network is sent to the second-layer network to obtain the second comprehensive feature; the second-layer network includes a fully connected layer and several convolutional layers to integrate and abstract the features;

[0086] Optionally, after weighted averaging the first comprehensive features A, B, and C, input them into the second-layer network to obtain the second comprehensive feature;

[0087] S33. The second comprehensive feature is used in the softmax layer to generate the predicted blood vessel segmentation result;

[0088] Optionally, perform weighted averaging on the second comprehensive features and the blood vessel segmentation results obtained from the first comprehensive features A, B, and C respectively, to make the blood vessel model smoother;

[0089] Furthermore, analyze the tumor angiogenesis status based on the comprehensive feature vector to obtain the analysis result, where the analysis result includes: evaluation of blood vessel density, blood vessel structure, blood vessel function, blood vessel inner diameter, carotid intima-media thickness, and atherosclerotic plaque location.

[0090] Furthermore, evolve the three-dimensional model of tumor blood vessels according to the comprehensive feature vector, and fuse the analysis result with the three-dimensional model of tumor blood vessels, including:

[0091] Obtain the blood vessel segmentation result according to the comprehensive feature vector, generate the three-dimensional model of blood vessels in the tumor region based on the blood vessel segmentation result, and perform color mapping on the analysis result, so that the three-dimensional model of blood vessels in the tumor region presents different colors.

[0092] Furthermore, evaluate the tumor angiogenesis status, including randomly selecting the coordinate values and pixel values of several feature points on the three-dimensional model of blood vessels in the tumor region, and the coordinate values and pixel values of the corresponding points on the real scan image; calculating the cumulative value of the pixel value deviation and the cumulative value of the coordinate value deviation; the feature points are the arc protruding points or end position points of the blood vessels.

[0093] Furthermore, the evaluation value is: the weighted value of the cumulative value of the pixel value deviation and the cumulative value of the coordinate value deviation.

[0094] Furthermore, when the cumulative value of the pixel value deviation, the cumulative value of the coordinate value deviation, and the weighted value of the cumulative value of the pixel value deviation and the cumulative value of the coordinate value deviation are all less than the threshold, the model modeling is qualified.

[0095] The tumor angiogenesis detection and analysis system based on multi-modal image fusion is used to execute the tumor angiogenesis detection and analysis method based on multi-modal image fusion, including:

[0096] An image acquisition module, used to acquire multi-modal image data of the target tumor;

[0097] A feature extraction module, used to extract features from the acquired multi-modal image data according to the neural network model;

[0098] An image fusion module, which is used to fuse the extracted features through a fusion network to obtain a comprehensive feature vector;

[0099] An image analysis module, which is used to analyze the tumor angiogenesis status according to the comprehensive feature vector to obtain an analysis result;

[0100] A model generation module, which is used to evolve a three-dimensional model of tumor blood vessels according to the comprehensive feature vector and fuse the analysis result with the three-dimensional model of tumor blood vessels;

[0101] A report generation module, which is used to evaluate the angiogenesis status of a tumor according to the analysis result and the fused three-dimensional model and generate a report.

[0102] Among them, segmenting the multi-modal image data of the target tumor collected includes: segmenting the region of the target tumor based on a conditional loss function;

[0103] The loss function model is:

[0104] ;

[0105] Among them, l ct represents the real tumor image, m ct represents the real tumor segmentation label, l p represents the pseudo-liver image generated by the generator, m p and m ct respectively represent the tumor segmentation label predicted based on the vascular structure and the real tumor segmentation label, m pi and m cti respectively represent the i-th tumor segmentation label predicted based on the vascular structure and the real tumor segmentation label; represents the absolute value of the difference between the real tumor image and the pseudo-tumor image calculated pixel by pixel, H, W, and D respectively represent the height, width, and depth of the tumor image, represents the pixel value of the real tumor image, represents the pixel value of the pseudo-tumor image, represents the binary cross-entropy loss, represents the DSC loss, Ls represents the image loss, and N is the number of images for data acquisition, 、 , respectively represent weight values.

[0106] It can provide more comprehensive tumor angiogenesis information and improve the accuracy of detection and analysis. Through the fusion of multi-modal image data, it can more realistically simulate the in-vivo tumor microenvironment. It can be used as a preclinical research platform for anti-tumor and anti-tumor angiogenesis drugs to provide support for drug screening and evaluation.

[0107] The technical features of the above embodiments can be combined arbitrarily (as long as there is no contradiction in the combination of these technical features). For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described; these embodiments not explicitly written should also be considered to be within the scope described in this specification.

[0108] In the foregoing, the present application has been described in a relatively specific and detailed manner through general descriptions and specific embodiments. It should be understood that based on the technical concept of the present application, several conventional adjustments or further innovations can be made to these specific embodiments; but as long as they do not depart from the technical concept of the present application, the technical solutions obtained by these conventional adjustments or further innovations also fall within the protection scope of the claims of the present application.

Claims

1. A method for detecting and analyzing tumor angiogenesis based on multimodal image fusion, characterized in that, Including: S1. Collect multi-modal image data of the target tumor; S2. Extract features from the collected multi-modal image data according to the neural network model; S3. Fuse the extracted features through a fusion network to obtain a comprehensive feature vector; The fusion network includes a first-layer network and a second-layer network. The first-layer network fuses the features of different modalities pairwise from micro to macro changes, and the second-layer network fuses the pairwise fused features to obtain a first comprehensive feature and a second comprehensive feature; Among them, obtaining the first comprehensive feature includes: S31. In the first-layer network, splice the feature vectors of the MRI tumor image and the PET tumor image to obtain a first comprehensive feature A, In the first-layer network, splice the feature vectors of the MRI tumor image and the CT tumor image to obtain a first comprehensive feature B, In the first-layer network, splice the feature vectors of the PET tumor image and the CT tumor image to obtain a first comprehensive feature C, S32. The comprehensive feature vector output by the first-layer network is sent to the second-layer network to obtain a second comprehensive feature; the second-layer network includes a fully connected layer and several convolutional layers to integrate and abstract features; After performing weighted averaging on the first comprehensive features A, B, and C, input them into the second-layer network to obtain a second comprehensive feature; S33. The second comprehensive feature is used in the softmax layer to generate a predicted blood vessel segmentation result; S4. Analyze the tumor angiogenesis state according to the comprehensive feature vector to obtain an analysis result; the analysis result includes: evaluation of blood vessel density, blood vessel structure, blood vessel function, blood vessel inner diameter, carotid intima-media thickness, and atherosclerotic plaque location; S5. Evolve the three-dimensional model of the tumor blood vessels according to the comprehensive feature vector and fuse the analysis result with the three-dimensional model of the tumor blood vessels; S6. Evaluate the angiogenesis state of the tumor according to the analysis result and the fused three-dimensional model, and generate a report.

2. The tumor angiogenesis detection and analysis method based on multimodal image fusion according to claim 1, characterized in that: Segment the multi-modal image data of the collected target tumor, including: segmenting the region of the target tumor based on a conditional loss function; The loss function model is: Among them, l ct represents the real tumor image, m ct represents the real tumor segmentation label, l p represents the pseudo-liver image generated by the generator, m p and m ct represent the tumor segmentation label predicted based on the vascular structure and the real tumor segmentation label respectively; represents the absolute value of the per-pixel difference between the real tumor image and the pseudo-tumor image. H, W, and K represent the height, width, and depth of the tumor image respectively, represents the pixel value of the real tumor image and represents the pixel value of the pseudo-tumor image, represents the binary cross-entropy loss, represents the DSC loss, 、 , represent the weight values respectively.

3. The tumor angiogenesis detection and analysis method based on multimodal image fusion according to claim 1, wherein: Extract features from the segmented multi-modal image data according to the neural network model, including the tumor image matching deformation process and the neural network model feature extraction process.

4. The method for detecting and analyzing tumor angiogenesis based on multimodal image fusion according to claim 3, wherein: The tumor image matching deformation process includes: S21. First, extract key points from the input image. The key points include corner points, edges, and textures; S22. After the key points, match the key points in the original image and the target image; S23. After the key point matching is completed, calculate the deformation matrix between the original image and the target image; the deformation matrix represents the distance and direction that each pixel point needs to move to align the original image with the target image; S24. Perform matching deformation on the source image according to the calculated deformation matrix; Where the original image is the MRI tumor image and the PET tumor image, and the target image is the CT tumor image.

5. The method for detecting and analyzing tumor angiogenesis based on multimodal image fusion according to claim 3, wherein The neural network model feature extraction process includes: The convolutional neural network with the AlexNet structure is used to extract the image features of MRI tumor images, PET tumor images, and CT tumor images. The 3D convolutional kernel is set to move sequentially at a preset step size, and a dot product operation is performed on the pixel values of the images at the corresponding covered positions. The result corresponds to a pixel point in the output image, and multiple pixel points after the operation form a feature map. A pooling layer is adopted to perform average pooling on the feature map to obtain a target feature map, and a feature vector is obtained through the target feature map, where the feature vector is obtained from the feature map through a flattening operation.

6. The method for detecting and analyzing tumor angiogenesis based on multimodal image fusion according to claim 1, wherein: The tumor angiogenesis state is analyzed based on the comprehensive feature vector to obtain an analysis result, and the analysis result includes: the evaluation of blood vessel density, blood vessel structure, blood vessel function, blood vessel inner diameter, carotid intima-media thickness, and the location of atherosclerotic plaques.

7. The method for detecting and analyzing tumor angiogenesis based on multimodal image fusion according to claim 1, wherein: The three-dimensional model of tumor blood vessels is evolved based on the comprehensive feature vector, and the analysis result is fused with the three-dimensional model of tumor blood vessels, including: The blood vessel segmentation result is obtained according to the comprehensive feature vector, the three-dimensional model of blood vessels in the tumor area is generated according to the blood vessel segmentation result, and the analysis result is color-mapped so that the three-dimensional model of blood vessels in the tumor area presents different colors.

8. A tumor angiogenesis detection and analysis system based on multimodal image fusion, which is used to execute the tumor angiogenesis detection and analysis method based on multimodal image fusion according to any one of claims 1-7, and is characterized in that: An image acquisition module, which is used to acquire multimodal image data of a target tumor; A feature extraction module, which is used to extract features from the acquired multimodal image data according to a neural network model; An image fusion module, which is used to fuse the extracted features through a fusion network to obtain a comprehensive feature vector; An image analysis module, which is used to analyze the tumor angiogenesis state according to the comprehensive feature vector to obtain an analysis result; A model generation module, which is used to evolve the three-dimensional model of tumor blood vessels according to the comprehensive feature vector and fuse the analysis result with the three-dimensional model of tumor blood vessels; A report generation module, which is used to evaluate the angiogenesis state of the tumor according to the analysis result and the fused three-dimensional model and generate a report; Among them, the segmentation of the acquired multimodal image data of the target tumor includes: segmenting the area of the target tumor based on a conditional loss function; The loss function model is: Among them, l ct represents the real tumor image, m ct represents the real tumor segmentation label, l p represents the pseudo - liver image generated by the generator, m p and m ct represent the tumor segmentation label predicted based on the vascular structure and the real tumor segmentation label respectively; represents the absolute value of the per - pixel difference between the real tumor image and the pseudo - tumor image. H, W, and K represent the height, width, and depth of the tumor image respectively, represents the pixel value of the real tumor image and represents the pixel value of the pseudo - tumor image, represents the binary cross - entropy loss, represents the DSC loss, 、 , represent the weight values respectively.

Images