Liver enhanced CT feature extraction and fusion method based on vascular topological structure

Abstract

The invention discloses a liver enhanced CT (Computed Tomography) feature extraction and fusion method based on a vascular topological structure, which relates to the technical field of medical image processing, and comprises the following steps: acquiring a liver enhanced CT image, preprocessing the liver enhanced CT image, rigidly registering the liver enhanced CT image and then taking the same as input, and acquiring a time feature vector based on a time feature extraction model; partitioning the registered liver enhanced CT image according to a vascular topological structure to obtain a plurality of liver partitioned images; dividing the time feature vector according to the plurality of liver partition images to obtain a partition feature vector corresponding to each liver partition image, taking the partition feature vectors as input, and performing spatial feature extraction and fusion based on a spatial feature extraction model to obtain a time-spatial feature vector; and calculating the time-space feature vector to obtain a classification result, and the like. According to the method, finer time and space feature extraction fusion of the liver enhanced CT image is realized, and the liver cancer detection and classification precision is improved.

Description

Technical Field

[0001] The present invention relates to a liver enhanced CT feature extraction and fusion method based on vascular topology, and belongs to the technical field of medical image processing. Background Art

[0002] Liver cancer is one of the malignant tumors with the highest morbidity and mortality rates worldwide. CT, especially enhanced CT, enhances image contrast by injecting contrast agents and has important diagnostic value in the early detection of liver cancer and lesion assessment.

[0003] The early symptoms of liver cancer are often not obvious, resulting in many patients being in the middle or late stages of the disease by the time they are diagnosed. CT (computed tomography), as an important tool in modern medical imaging technology, plays a key role in the diagnosis of liver cancer. CT relies on natural differences in tissue absorption of X-rays to generate images and is commonly used to evaluate bone structure, lung diseases, acute conditions (such as trauma), etc. Enhanced CT is a CT scanning method that enhances the contrast of tissue and organ images by injecting a contrast agent (usually an iodine-containing contrast agent). It can better detect and evaluate the characteristics of diseases such as tumors, vascular lesions, and organ lesions, and has stronger diagnostic capabilities. It is commonly used to observe lesions in organs such as blood vessels, tumors, and the liver.

[0004] Computer-aided diagnosis methods based on CT images and deep learning have effectively improved the accuracy of doctors' diagnoses. However, existing methods still have shortcomings in image temporal feature extraction and spatial feature fusion. Specifically, existing methods usually only focus on CT image feature extraction in a single phase, such as the arterial phase, venous phase, or excretion phase. It is difficult to fully capture the dynamic evolution of the tumor, and often ignores the complex topological structure of the liver, resulting in limited accuracy in liver cancer diagnosis and classification. In view of this, the present invention proposes a liver partitioning method and integrates multi-period feature extraction and fusion technology, which is of great significance to improving the accuracy of liver cancer detection and classification. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method for liver enhanced CT feature extraction and fusion based on vascular topology to achieve more refined temporal and spatial feature extraction and fusion of liver enhanced CT images, thereby improving the accuracy of liver cancer detection and classification.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0007] The present invention provides a liver enhanced CT feature extraction and fusion method based on vascular topology, comprising:

[0008] Acquiring liver enhanced CT images, wherein the liver enhanced CT images are liver enhanced CT images during arterial phase, venous phase, delayed phase, and plain scan phase;

[0009] preprocessing the liver enhanced CT image to obtain a preprocessed liver enhanced CT image;

[0010] Performing rigid registration on the pre-processed liver enhanced CT image to obtain a registered liver enhanced CT image;

[0011] The registered liver enhanced CT image is used as input, and the temporal feature extraction is performed based on the temporal feature extraction model to obtain the temporal feature vector;

[0012] The registered liver enhanced CT image is partitioned according to the vascular topology to obtain multiple liver partition images;

[0013] Dividing the time feature vector according to multiple liver partition images to obtain partition feature vectors corresponding to each liver partition image;

[0014] The partition feature vectors corresponding to each liver partition image are used as input, and spatial feature extraction and fusion are performed based on the spatial feature extraction model to obtain the temporal and spatial feature vectors;

[0015] The classification results are obtained based on the time-space feature vectors.

[0016] Furthermore, the liver enhanced CT images all use the same CT parameters, which include slice thickness and scanning voltage.

[0017] Furthermore, the preprocessing of the liver enhanced CT image to obtain the preprocessed liver enhanced CT image includes:

[0018] Performing Gaussian filtering on the liver enhanced CT image to obtain a Gaussian filtered liver enhanced CT image;

[0019] The liver enhanced CT images after Gaussian filtering are spatially resampled to be unified to a fixed resolution to obtain spatially resampled liver enhanced CT images, thereby obtaining preprocessed liver enhanced CT images.

[0020] Furthermore, performing rigid registration on the preprocessed liver-enhanced CT image to obtain a registered liver-enhanced CT image includes:

[0021] The liver enhanced CT image in the venous phase is used as the target image, and the liver enhanced CT images in the arterial phase and delayed phase are used as the original images for rigid registration. During the rigid registration process, the objective function is gradually optimized using the gradient descent method to minimize the similarity measure between the original image and the target image to obtain the registered arterial phase image and the registered delayed phase image.

[0022] The venous phase liver enhanced CT image, the registered arterial phase image, the registered delayed phase image, and the plain scan phase liver enhanced CT image constitute a registered liver enhanced CT image;

[0023] The expression of the rigid registration is:

[0024] ,

[0025] in, represents the image coordinates after rigid registration, represents the rotation matrix, represents the image coordinates before rigid registration, represents the translation vector;

[0026] The expression of the objective function is:

[0027] ,

[0028] in, represents the similarity measure, Represents the total number of pixels in the image. Indicates the position of the image before rigid registration The pixel value at Indicates that the image after rigid registration is at position The pixel value at .

[0029] Furthermore, partitioning the registered liver enhanced CT image according to the vascular topology to obtain a plurality of liver partition images includes:

[0030] Performing data cleaning on the registered liver enhanced CT image to obtain a data-cleaned liver enhanced CT image;

[0031] The liver enhanced CT image after data cleaning is used as input, and the liver and blood vessel regions are automatically segmented based on the U-shaped network to obtain the segmented liver enhanced CT image;

[0032] Calculating a distance transformation matrix of the portal vein according to the segmented liver enhanced CT image, wherein the set of pixels with the largest distance transformation in the distance transformation matrix is ​​the centerline of the vein;

[0033] Traverse each pixel point on the centerline of the blood vessel. If the neighborhood of a certain pixel point includes multiple branches, it is judged as a branch point. All pixels between two adjacent branch points constitute a blood vessel segment, forming a blood vessel topology map;

[0034] Traverse each pixel point in the liver enhanced CT image after data cleaning, calculate the Euclidean distance between each pixel point and each vascular segment, and assign it to the nearest vascular segment to form multiple liver partition images.

[0035] Furthermore, the expression for calculating the distance transformation matrix of the portal vein according to the segmented liver enhanced CT image is:

[0036] ,

[0037] in, represents the distance transformation matrix of the portal vein, Represents the pixel coordinates, represents the coordinates of the blood vessel boundary, represents the set of blood vessel boundaries;

[0038] The expression of the blood vessel centerline is:

[0039] ,

[0040] in, Indicates the centerline of the blood vessel;

[0041] The expression of the blood vessel segment is:

[0042] ,

[0043] in, represents the kth blood vessel segment, represents the i-th pixel point on the centerline of the blood vessel, represents the j-th branch point, represents the j+1th branch point;

[0044] The expression of the liver partition image is:

[0045] ,

[0046] in, Represents pixel points The liver area image to which it belongs, Represents pixel points coordinate, Indicates the distance pixel point on the i-th blood vessel segment The coordinates of the nearest pixel, represents the i-th blood vessel segment, Represents a collection of vessel segments.

[0047] Furthermore, the temporal features are divided according to the multiple liver partition images to obtain partition feature vectors corresponding to the respective liver partition images, and the vector lengths are unified by average pooling.

[0048] Furthermore, the temporal feature extraction model adopts a residual network, which includes an input layer, a residual block and an output layer connected in sequence;

[0049] The input layer is provided with 4 input channels, each of which corresponds to the venous phase, arterial phase, delayed phase and plain scan phase in the registered liver enhanced CT image;

[0050] There are multiple residual blocks, and their processing expressions are:

[0051] ,

[0052] in, represents the output features, represents the input features, represents the residual function, Represents the weight parameter of the residual function;

[0053] The spatial feature extraction model includes an input layer, a word embedding layer, an encoder, a Transformer layer, a decoder, an output embedding layer and a position encoding layer connected in sequence, wherein the number of the Transformer layers is 10.

[0054] Furthermore, the classification result obtained according to the time-space feature vector is achieved through a fully connected network.

[0055] Furthermore, the method further includes pre-training the temporal feature extraction model, the spatial feature extraction model, and the fully connected network, wherein the pre-training method includes:

[0056] Acquire liver enhanced CT image dataset;

[0057] Annotate the liver, blood vessels, and tumor regions of the liver enhanced CT image dataset to obtain an annotated liver enhanced CT image dataset;

[0058] The labeled liver enhanced CT image dataset is divided into a training set and a test set;

[0059] The training set data is used as input to train the spatial feature extraction model and the fully connected network, and the cross entropy loss function is used to adjust the model parameters during the training process;

[0060] The test set data is used as input to test the spatial feature extraction model and the fully connected network. Evaluation indicators are calculated based on the test results, and the pre-trained spatial feature extraction model and the fully connected network are obtained based on the evaluation indicators.

[0061] The evaluation indicators include accuracy, recall and F1-score.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] The present invention uses a deep learning model to extract the temporal features of images in each period from the registered images, enabling the model to more fully learn the complex patterns of liver lesions. This method can more comprehensively capture the key information in liver-enhanced CT images, avoiding the limitations of single-dimensional analysis, and performing spatial feature fusion based on vascular topology and the Transformer model to achieve more refined spatial feature fusion and improve the accuracy of liver cancer detection and classification. The feature extraction and fusion method for liver-enhanced CT images of the present invention not only conforms to the physiological characteristics of the liver, but also improves the pertinence and accuracy of feature extraction, making the analysis results more biologically meaningful. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 Schematic diagram of a flow chart of a liver enhanced CT feature extraction and fusion method based on vascular topology in one embodiment of the present invention;

[0065] Figure 2 Schematic diagram of the structure of a temporal feature extraction model of a liver enhanced CT feature extraction and fusion method based on vascular topology in one embodiment of the present invention;

[0066] Figure 3 A schematic diagram of segmentation and partitioning of a liver enhanced CT feature extraction and fusion method based on vascular topology in an embodiment of the present invention;

[0067] Figure 4 Schematic diagram of the structure of a spatial feature extraction model of a liver enhanced CT feature extraction and fusion method based on vascular topology in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0069] Example 1:

[0070] like Figure 1 As shown, an embodiment of the present invention provides a liver enhanced CT feature extraction and fusion method based on vascular topology, comprising the following steps:

[0071] Obtain liver contrast-enhanced CT images, including three-phase liver contrast-enhanced CT images and plain-contrast liver contrast-enhanced CT images (arterial, venous, and delayed phases), to comprehensively reflect the morphology and function of the liver and blood vessels. It is important to note that all liver contrast-enhanced CT images must use uniform CT parameters, including slice thickness (e.g., 1mm or 5mm) and scanning voltage (e.g., 120kVp).

[0072] Preprocessing of liver enhanced CT images includes:

[0073] The OpenCV library in Python was used to perform Gaussian filtering on the CT image data to smooth the image and reduce random noise interference. The SimpleITK library was then used to spatially resample the data to a fixed resolution to obtain preprocessed liver enhanced CT images.

[0074] Next, the SimpleITK tool was used to perform rigid registration on the preprocessed liver enhancement CT images. The venous phase liver enhancement CT images were used as the target images, and the arterial phase and delayed phase liver enhancement CT images were used as the original images for rigid registration. The expression for rigid registration is:

[0075] ,

[0076] in, represents the image coordinates after rigid registration, represents the rotation matrix, represents the image coordinates before rigid registration, Represents the translation vector.

[0077] In the rigid registration process, the objective function is gradually optimized using the gradient descent method to minimize the similarity measure between the original image and the target image to obtain the registered arterial phase image and the registered delayed phase image. The expression of the objective function is:

[0078] ,

[0079] in, represents the similarity measure, Represents the total number of pixels in the image. Indicates the position of the image before rigid registration The pixel value at Indicates that the image after rigid registration is at position The pixel value at .

[0080] The liver enhanced CT image in the venous phase, the registered arterial phase image, the registered delayed phase image, and the liver enhanced CT image in the plain scan phase constitute the registered liver enhanced CT image.

[0081] The registered liver enhanced CT image is used as input. Specifically, the venous phase liver enhanced CT image, the registered arterial phase image, the registered delayed phase image, and the plain scan liver enhanced CT image are respectively used as a channel input. Time feature extraction is performed based on the time feature extraction model to obtain a time feature vector.

[0082] In this embodiment, the temporal feature extraction model adopts a residual network model, such as Figure 2 As shown in Figure 1, the temporal feature extraction model includes an input layer, a residual block, and an output layer connected in sequence. There are five residual blocks, and the data processing expression is:

[0083] ,

[0084] in, represents the output features, represents the input features, represents the residual function, Represents the weight parameter of the residual function.

[0085] Combine Figure 3 , the registered liver enhanced CT image is partitioned according to the vascular topology to obtain multiple liver partition images, specifically:

[0086] First, the registered liver enhanced CT images were cleaned to remove tiny segmentation noise to obtain the cleaned liver enhanced CT images. Then, the U-net was used to automatically segment the liver and vascular regions to obtain the segmented liver enhanced CT images.

[0087] The distance transformation matrix of the portal vein is calculated using the skimage library for the segmented liver enhanced CT image. The expression is:

[0088] ,

[0089] in, represents the distance transformation matrix of the portal vein, Represents the pixel coordinates, represents the coordinates of the blood vessel boundary, Represents a collection of blood vessel boundaries.

[0090] The point set consisting of the pixel points with the largest distance transformation in the distance transformation matrix is ​​taken as the blood vessel centerline. The expression of the blood vessel centerline is:

[0091] ,

[0092] in, Indicates the centerline of the blood vessel.

[0093] Traverse each pixel point on the centerline of the blood vessel. If the neighborhood of a pixel point includes multiple branches, it is considered a branch point. All pixels between two adjacent branch points form a blood vessel segment, forming a blood vessel topology diagram. The expression of the blood vessel segment is:

[0094] ,

[0095] in, represents the kth blood vessel segment, represents the i-th pixel point on the centerline of the blood vessel, represents the j-th branch point, Indicates the j+1th branch point.

[0096] Traverse each pixel point in the liver enhanced CT image after data cleaning, calculate the Euclidean distance between each pixel point and each vascular segment, and assign it to the vascular segment closest to it to form multiple liver partition images. The expression of the liver partition image is:

[0097] ,

[0098] in, Represents pixel points The liver area image to which it belongs, Represents pixel points coordinate, Indicates the distance pixel point on the i-th blood vessel segment The coordinates of the nearest pixel, represents the i-th blood vessel segment, Represents a collection of vessel segments.

[0099] The temporal features are divided according to multiple liver partition images to obtain the partition feature vectors corresponding to each liver partition image, and the dimensions of each partition feature vector are made consistent through average pooling. Then, multiple partition feature vectors are used as input, and spatial feature extraction is performed based on the spatial feature extraction model to obtain the temporal spatial feature vector.

[0100] like Figure 4 As shown, the spatial feature extraction model includes an input layer, a word embedding layer, an encoder, a Transformer layer, a decoder, an output embedding layer, and a position encoding layer connected in sequence. In this embodiment, the number of Transformer layers is 10.

[0101] Finally, the time-space feature vectors are passed through a fully connected network to obtain the classification results, which are then used as a reference for medical staff to diagnose whether the patient has liver cancer.

[0102] Example 2:

[0103] Based on Example 1, this embodiment further includes pre-training the temporal feature extraction model, the spatial feature extraction model, and the fully connected network. The pre-training method includes:

[0104] Obtain liver enhanced CT image dataset.

[0105] ITK-SNAP software was used to annotate the liver, blood vessels, and tumor areas in the liver enhanced CT image dataset.

[0106] The training set data is used as input to train the temporal feature extraction model, spatial feature extraction model and fully connected network together, and the cross entropy loss function is used to adjust the model parameters during the training process.

[0107] The test set data is used as input to test the temporal feature extraction model, spatial feature extraction model and fully connected network together. Evaluation indicators are calculated based on the test results, including accuracy, recall rate and F1-score, and the pre-trained spatial feature extraction model and fully connected network are obtained based on the evaluation indicators.

[0108] The classification ability and robustness of the model were evaluated by drawing the ROC curve and calculating the area under the line (AUC), and the generalization ability of the model to different data distributions was ensured through cross-validation.

[0109] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0110] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0111] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0112] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0113] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which are all protected by the present invention.

Claims

1. A liver enhanced CT feature extraction and fusion method based on vascular topology, characterized in that: include: Acquiring liver enhanced CT images, wherein the liver enhanced CT images include liver enhanced CT images in arterial phase, venous phase, delayed phase and plain scan phase; Preprocessing the liver enhanced CT image to obtain a preprocessed liver enhanced CT image; Performing rigid registration on the preprocessed liver enhanced CT image to obtain the registered liver enhanced CT image; The registered liver enhanced CT image is used as input, and the temporal feature extraction is performed based on the temporal feature extraction model to obtain the temporal feature vector; The registered liver enhanced CT image is segmented and partitioned according to the vascular topology to obtain multiple liver partition images; Dividing the time feature vector according to multiple liver partition images to obtain partition feature vectors corresponding to each liver partition image; The partition feature vectors corresponding to each liver partition image are used as input, and spatial feature extraction and fusion are performed based on the spatial feature extraction model to obtain the temporal and spatial feature vectors; The classification results are obtained based on the time-space feature vectors.

2. The method for extracting and fusion features of liver enhanced CT based on vascular topology according to claim 1, characterized in that: The liver enhanced CT images all use the same CT parameters, which include layer thickness and scanning voltage.

3. The method for extracting and fusion features of liver enhanced CT based on vascular topology according to claim 1, characterized in that: The preprocessing of the liver enhanced CT image to obtain the preprocessed liver enhanced CT image includes: Performing Gaussian filtering on the liver enhanced CT image to obtain the liver enhanced CT image after Gaussian filtering; The liver enhanced CT images after Gaussian filtering were spatially resampled to be unified to a fixed resolution to obtain preprocessed liver enhanced CT images.

4. The method for extracting and fusion features of liver enhanced CT based on vascular topology according to claim 1, characterized in that: The step of rigidly registering the preprocessed liver enhanced CT image to obtain the registered liver enhanced CT image comprises: The liver enhanced CT images in the venous phase are taken as the target images, and the liver enhanced CT images in the arterial phase and delayed phase are taken as the original images for rigid registration. In the rigid registration process, the objective function is gradually optimized by the gradient descent method to minimize the similarity measure between the original image and the target image, so as to obtain the registered arterial phase image and the registered delayed phase image. The liver enhanced CT image in the venous phase, the registered arterial phase image, the registered delayed phase image, and the liver enhanced CT image in the plain scan phase constitute a registered liver enhanced CT image; The expression of the rigid registration is: ， in, represents the image coordinates after rigid registration, represents the rotation matrix, represents the image coordinates before rigid registration, represents the translation vector; The expression of the objective function is: ， in, represents the similarity measure, Represents the total number of pixels in the image. Indicates the position of the image before rigid registration The pixel value at Indicates the position of the image after rigid registration The pixel value at .

5. The method for extracting and fusion features of liver enhanced CT based on vascular topology according to claim 1, characterized in that: The step of partitioning the registered liver enhanced CT image according to the vascular topology structure to obtain a plurality of liver partition images includes: The registered liver enhanced CT image is data cleaned to obtain a data cleaned liver enhanced CT image; the data cleaned liver enhanced CT image is used as input, and the liver and blood vessel regions are automatically segmented based on a U-shaped network to obtain a segmented liver enhanced CT image; Calculate the distance transformation matrix of the portal vein according to the segmented liver enhanced CT image, and use the point set consisting of the pixel points with the largest distance transformation in the distance transformation matrix as the centerline of the blood vessel; Traverse each pixel point on the center line of the blood vessel. If the neighborhood of a pixel point includes multiple branches, it is judged as a branch point. All the pixels between two adjacent branch points constitute a blood vessel segment to form a blood vessel topology diagram. Traverse each pixel point in the liver enhanced CT image after data cleaning, calculate the Euclidean distance between each pixel point and each vascular segment, and attribute it to the vascular segment closest to it, forming multiple liver partition images.

6. The method for extracting and fusion features of liver enhanced CT based on vascular topology according to claim 5, characterized in that: The expression for calculating the distance transformation matrix of the portal vein according to the segmented liver enhanced CT image is: ， in, represents the distance transformation matrix of the portal vein, Represents the pixel coordinates, represents the coordinates of the blood vessel boundary, represents the set of blood vessel boundaries; The expression of the blood vessel centerline is: ， in, Indicates the centerline of the blood vessel; The expression of the blood vessel segment is: ， in, represents the kth blood vessel segment, represents the i-th pixel point on the centerline of the blood vessel, represents the j-th branch point, represents the j+1th branch point; The expression of the liver partition image is: ， in, Represents pixel The liver area image to which it belongs, Represents pixel coordinate, Indicates the distance pixel point on the i-th blood vessel segment The coordinates of the nearest pixel. represents the i-th blood vessel segment, Represents a collection of vessel segments.

7. The method for extracting and fusion features of liver enhanced CT based on vascular topology according to claim 1, characterized in that: The temporal features are divided according to a plurality of liver partition images to obtain partition feature vectors corresponding to the respective liver partition images, and the vector lengths are unified through average pooling.

8. The method for extracting and fusion features of liver enhanced CT based on vascular topology according to claim 1, characterized in that: The temporal feature extraction model adopts a residual network, which includes an input layer, a residual block and an output layer connected in sequence; The input layer is provided with 4 input channels, each of which corresponds to the venous phase, arterial phase, delayed phase and plain scan phase in the registered liver enhanced CT image; There are multiple residual blocks, and their processing expressions are: ， in, represents the output features, represents the input features, represents the residual function, Represents the weight parameter of the residual function; The spatial feature extraction model includes an input layer, a word embedding layer, an encoder, a Transformer layer, a decoder, an output embedding layer and a position encoding layer connected in sequence, wherein the number of the Transformer layers is 10.

9. The method for extracting and fusion features of liver enhanced CT based on vascular topology according to claim 1, characterized in that: The classification result obtained according to the time-space feature vector is achieved through a fully connected network.

10. The method for extracting and fusion features of liver enhanced CT based on vascular topology according to claim 1, characterized in that: It also includes pre-training the temporal feature extraction model, the spatial feature extraction model and the fully connected network, wherein the pre-training method includes: Acquire liver enhanced CT image dataset; Annotate the liver, blood vessels and tumor regions of the liver enhanced CT image dataset to obtain annotated liver enhanced CT image dataset; The labeled liver enhanced CT image dataset is divided into a training set and a test set; The training set data is used as input to train the temporal feature extraction model, the spatial feature extraction model and the fully connected network, and the cross entropy loss function is used to adjust the model parameters during the training process; The test set data is used as input to test the spatial feature extraction model and the fully connected network, and the evaluation index is calculated according to the test results. Based on the evaluation index, the pre-trained temporal feature extraction model, spatial feature extraction model and fully connected network are obtained; The evaluation indicators include accuracy, recall and F1-score.

Images