A Feature Extraction and Fusion Method for Enhanced CT of Liver Based on Vascular Topology Structure

Abstract

The present invention discloses a method for extracting and fusing liver enhanced CT features based on vascular topological structure, which relates to the technical field of medical image processing. The method includes steps of obtaining liver enhanced CT images, preprocessing them, and using the preprocessed and rigidly registered images as input to obtain time feature vectors based on a time feature extraction model; partitioning the registered liver enhanced CT images according to the vascular topological structure to obtain multiple liver partition images; dividing the time feature vectors according to the multiple liver partition images to obtain partition feature vectors corresponding to each liver partition image, and using them as input to perform spatial feature extraction and fusion based on a spatial feature extraction model to obtain time-space feature vectors; calculating the time-space feature vectors to obtain classification results, etc. The present invention realizes more refined feature extraction and fusion of liver enhanced CT images in terms of time and space, and improves the detection and classification accuracy of liver cancer.

Description

Technical Field

[0001] The present invention relates to a method for extracting and fusing enhanced CT features of the liver based on vascular topology, belonging to the technical field of medical image processing. Background Art

[0002] Liver cancer is one of the malignant tumors with relatively high incidence and mortality globally. CT, especially enhanced CT, by injecting a contrast agent to enhance the image contrast, has important diagnostic value in the early detection and lesion evaluation of liver cancer.

[0003] The early symptoms of liver cancer are usually not obvious, resulting in many patients being in the middle or late stage of the disease at the time of diagnosis. CT (i.e., computed tomography), as an important tool of modern medical imaging technology, plays a key role in the diagnosis of liver cancer. CT relies on the natural differences in tissue absorption of X-rays to generate images, and is usually used to evaluate bone structures, lung diseases, acute conditions (such as trauma), etc. Enhanced CT is a CT scanning method that enhances the image contrast of tissues and organs by injecting a contrast agent (usually an iodine-containing contrast agent), which can better detect and evaluate the characteristics of diseases such as tumors, vascular lesions, and organ lesions, has stronger diagnostic capabilities, and is usually used to observe the lesions of 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 deficiencies in image temporal feature extraction and spatial feature fusion. Specifically, existing methods usually only focus on the feature extraction of CT images in a single phase such as the arterial phase, venous phase, or excretory phase, making it difficult to fully capture the dynamic evolution process of tumors, and often ignoring the complex topological structure of the liver, resulting in limited accuracy in the diagnosis and classification of liver cancer. In view of this, the present invention proposes a liver partitioning method and fuses multi-phase feature extraction and fusion technologies, which is of great significance for improving the detection and classification accuracy of liver cancer. 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 extracting and fusing enhanced CT features of the liver based on vascular topology, so as to achieve more refined extraction and fusion of temporal and spatial features of enhanced CT images of the liver and improve the detection and classification accuracy of liver cancer.

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

[0007] The present invention provides a method for extracting and fusing enhanced CT features of the liver based on vascular topology, including:

[0008] Obtain enhanced CT images of the liver, where the enhanced CT images of the liver are enhanced CT of the arterial phase, venous phase, delayed phase, and plain scan phase;

[0009] Preprocess the enhanced CT images of the liver to obtain preprocessed enhanced CT images of the liver;

[0010] Perform rigid registration on the preprocessed enhanced CT images of the liver to obtain registered enhanced CT images of the liver;

[0011] Take the registered enhanced CT images of the liver as input, and perform temporal feature extraction based on a temporal feature extraction model to obtain a temporal feature vector;

[0012] Partition the registered enhanced CT images of the liver according to the vascular topological structure to obtain multiple liver partition images;

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

[0014] Take the partition feature vectors corresponding to each liver partition image as input, and perform spatial feature extraction and fusion based on a spatial feature extraction model to obtain a spatio-temporal feature vector;

[0015] Obtain a classification result according to the spatio-temporal feature vector.

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

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

[0018] Perform Gaussian filtering on the enhanced CT images of the liver to obtain enhanced CT images of the liver after Gaussian filtering;

[0019] Perform spatial resampling on the enhanced CT images of the liver after Gaussian filtering to uniformly convert them to a fixed resolution to obtain enhanced CT images of the liver after spatial resampling, and obtain preprocessed enhanced CT images of the liver.

[0020] Furthermore, the rigid registration of the preprocessed enhanced CT images of the liver to obtain registered enhanced CT images of the liver includes:

[0021] Take the enhanced CT image of the liver in the venous phase as the target image, and take the enhanced CT images of the liver in the arterial phase and the delayed phase as the original images respectively for rigid registration. During the rigid registration process, use the gradient descent method to gradually optimize the objective function to make the similarity metric value between the original image and the target image the smallest, and obtain the registered arterial phase image and the registered delayed phase image;

[0022] The registered contrast-enhanced CT images of the liver include the liver contrast-enhanced CT images in the venous phase, the registered arterial-phase images, the registered delayed-phase images, and the liver contrast-enhanced CT images in the non-contrast phase;

[0023] The expression of the rigid registration is:

[0024] ,

[0025] where, 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] where, represents the similarity metric, represents the total number of pixels in the image, represents the pixel value of the image before rigid registration at position , represents the pixel value of the image after rigid registration at position .

[0029] Furthermore, partitioning the registered contrast-enhanced CT images of the liver according to the vascular topology structure to obtain multiple liver partition images includes:

[0030] Performing data cleaning on the registered contrast-enhanced CT images of the liver to obtain the contrast-enhanced CT images of the liver after data cleaning;

[0031] Using the contrast-enhanced CT images of the liver after data cleaning as the input, automatically segmenting the liver and vascular regions based on the U-shaped network to obtain the segmented contrast-enhanced CT images of the liver;

[0032] Calculating the distance transformation matrix of the portal vein vessels according to the segmented contrast-enhanced CT images of the liver. The set of pixel points with the maximum distance transformation in the distance transformation matrix is the vascular centerline;

[0033] Traversing each pixel point on the vascular centerline. If a certain pixel point's neighborhood includes multiple branches, it is determined as a branch point. All pixel points between two adjacent branch points form a vascular segment, forming a vascular topology structure diagram;

[0034] Traversing each pixel point in the contrast-enhanced CT images of the liver after data cleaning, calculating the Euclidean distance of each pixel point from each vascular segment, and attributing it to the vascular segment with the closest distance, forming multiple liver partition images.

[0035] Furthermore, the expression for calculating the distance transformation matrix of the portal vein vessels based on the segmented enhanced CT images of the liver is as follows:

[0036] ,

[0037] where represents the distance transformation matrix of the portal vein vessels, represents the pixel coordinates, represents the vascular boundary coordinates, represents the set of vascular boundaries;

[0038] The expression for the vascular centerline is as follows:

[0039] ,

[0040] where represents the vascular centerline;

[0041] The expression for the vascular segment is as follows:

[0042] ,

[0043] where represents the k-th vascular segment, represents the i-th pixel on the vascular centerline, represents the j-th branch point, represents the (j + 1)-th branch point;

[0044] The expression for the liver partition image is as follows:

[0045] ,

[0046] where represents the liver partition image to which the pixel belongs, represents the pixel coordinates, represents the pixel coordinates of the pixel on the i-th vascular segment that is closest to the pixel , represents the i-th vascular segment, represents the set of vascular segments.

[0047] Furthermore, the partition feature vectors corresponding to the respective liver partition images obtained by dividing the temporal features based on multiple liver partition images are unified in vector length through average pooling.

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

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

[0050] A plurality of residual blocks are provided, and its processing expression is:

[0051] ,

[0052] wherein, represents the output feature, represents the input feature, 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. Among them, the number of the Transformer layers is 10.

[0054] Further, obtaining the classification result according to the spatio-temporal feature vector is realized through a fully connected network.

[0055] Further, it also includes pre-training the time feature extraction model, the spatial feature extraction model, and the fully connected network. The pre-training method includes:

[0056] Obtain a dataset of enhanced liver CT images;

[0057] Label the liver, blood vessels, and tumor regions in the dataset of enhanced liver CT images to obtain a labeled dataset of enhanced liver CT images;

[0058] Divide the labeled dataset of enhanced liver CT images into a training set and a test set;

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

[0060] Use the test set data as input to test the spatial feature extraction model and the fully connected network, calculate the evaluation index according to the test result, and obtain the pre-trained spatial feature extraction model and fully connected network based on the evaluation index;

[0061] Among them, the evaluation indexes include accuracy rate, recall rate, and F1-score.

[0062] Compared with the prior art, the beneficial effects achieved by the present invention:

[0063] The present invention extracts the temporal features of images in each period by using a deep learning model for 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, avoid the limitations of single-dimensional analysis, and perform spatial feature fusion based on the vascular topological structure 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 according to the present invention not only conforms to the physiological characteristics of the liver but also can improve the pertinence and accuracy of feature extraction, making the analysis results more biologically significant. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 FIG. is a schematic flowchart of a method for extracting and fusing liver enhanced CT features based on a vascular topological structure in an embodiment of the present invention;

[0065] Figure 2 FIG. is a schematic structural diagram of a temporal feature extraction model of a method for extracting and fusing liver enhanced CT features based on a vascular topological structure in an embodiment of the present invention;

[0066] Figure 3 FIG. is a schematic diagram of segmentation and zoning of a method for extracting and fusing liver enhanced CT features based on a vascular topological structure in an embodiment of the present invention;

[0067] Figure 4 FIG. is a schematic structural diagram of a spatial feature extraction model of a method for extracting and fusing liver enhanced CT features based on a vascular topological structure in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

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

[0069] Embodiment 1:

[0070] As Figure 1 shown, an embodiment of the present invention provides a method for extracting and fusing liver enhanced CT features based on a vascular topological structure, including the following steps:

[0071] Obtain liver enhanced CT images, which include three-phase liver enhanced CT images and a plain scan liver enhanced CT image. The three-phase liver enhanced CT images are the arterial phase, the venous phase, and the delayed phase, so as to comprehensively reflect the morphology and functional state of the liver and blood vessels. It should be noted that all liver enhanced CT images need to use unified CT parameters, and the CT parameters include slice thickness (such as 1 mm or 5 mm), scanning voltage (such as 120 kVp), etc.

[0072] Preprocess the enhanced CT images of the liver, specifically including:

[0073] Use the OpenCV library in Python to perform Gaussian filtering on the CT image data to smooth the image and reduce random noise interference. Then, use the SimpleITK library to perform spatial resampling on the data to unify it to a fixed resolution, obtaining the preprocessed enhanced CT images of the liver.

[0074] Next, use the SimpleITK tool to perform rigid registration on the preprocessed enhanced CT images of the liver. Take the enhanced CT image of the liver in the venous phase as the target image, and take the enhanced CT images of the liver in the arterial phase and the delayed phase as the original images respectively for rigid registration. The expression for rigid registration is:

[0075] ,

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

[0077] During the rigid registration process, use the gradient descent method to gradually optimize the objective function to minimize the similarity metric value between the original image and the target image, obtaining the registered arterial phase image and the registered delayed phase image. The expression for the objective function is:

[0078] ,

[0079] Among them, represents the similarity metric, represents the total number of pixel points in the image, represents the pixel value of the image before rigid registration at position , represents the pixel value of the image after rigid registration at position 。

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

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

[0082] In this embodiment, the time feature extraction model adopts a residual network model. As Figure 2 shown, the time feature extraction model includes an input layer, residual blocks, and an output layer connected in sequence. There are 5 residual blocks, and its data processing expression is:

[0083] ,

[0084] wherein, represents the output feature, represents the input feature, represents the residual function, represents the weight parameter of the residual function.

[0085] Combined with Figure 3 , multiple liver partition images are obtained by partitioning the registered contrast-enhanced CT image of the liver according to the vascular topological structure. Specifically:

[0086] First, the registered contrast-enhanced CT image of the liver is subjected to data cleaning to remove minute segmentation noise, obtaining the contrast-enhanced CT image of the liver after data cleaning. Then, a U-shaped network (U-net) is used to automatically segment the liver and vascular regions to obtain the segmented contrast-enhanced CT image of the liver.

[0087] The distance transformation matrix of the portal vein vessels is calculated for the segmented contrast-enhanced CT image of the liver using the skimage library, and its expression is:

[0088] ,

[0089] wherein, represents the distance transformation matrix of the portal vein vessels, represents the pixel point coordinates, represents the vascular boundary coordinates, represents the vascular boundary set.

[0090] The point set composed of the pixel points with the maximum distance transformation in the distance transformation matrix is used as the vascular centerline, and the expression of the vascular centerline is:

[0091] ,

[0092] wherein, represents the vascular centerline.

[0093] Traverse each pixel point on the vascular centerline. If the neighborhood of a certain pixel point includes multiple branches, it is determined as a branch point. All the pixel points between two adjacent branch points form a vascular segment, forming a vascular topological structure diagram. The expression of the vascular segment is:

[0094] ,

[0095] Among them, represents the k-th blood vessel segment, represents the i-th pixel point on the blood vessel center line, represents the j-th branch point, represents the (j + 1)-th branch point.

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

[0097] ,

[0098] Among them, represents the liver partition image to which the pixel point belongs, represents the coordinates of the pixel point , represents the coordinates of the pixel point on the i-th blood vessel segment that is closest to the pixel point , represents the i-th blood vessel segment, represents the set of blood vessel segments.

[0099] According to the multiple liver partition images, the time features are divided 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, taking the multiple partition feature vectors as inputs, spatial feature extraction is performed based on the spatial feature extraction model to obtain the spatio-temporal feature vectors.

[0100] As Figure 4 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 spatio-temporal feature vectors are passed through a fully connected network to obtain the classification result, which will be used as a reference for medical staff to diagnose whether the patient has liver cancer in the future.

[0102] Embodiment 2:

[0103] On the basis of Embodiment 1, this embodiment further includes jointly pre-training the time feature extraction model, the spatial feature extraction model, and the fully connected network. The pre-training method includes:

[0104] Obtain a dataset of enhanced CT images of the liver.

[0105] The ITK-SNAP software is used to annotate the liver, blood vessels, and tumor regions in the enhanced CT image dataset of the liver.

[0106] Taking the training set data as input, the time feature extraction model, the spatial feature extraction model, and the fully connected network are trained together, and the cross-entropy loss function is used to adjust the model parameters during the training process.

[0107] Taking the test set data as input, the time feature extraction model, the spatial feature extraction model, and the fully connected network are tested together. Evaluation metrics are calculated based on the test results. The evaluation metrics include accuracy, recall, F1-score, etc., and the pre-trained spatial feature extraction model and fully connected network are obtained based on the evaluation metrics.

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

[0109] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0110] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0111] These computer program instructions can 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, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0112] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are executed on the computer or other programmable apparatus to generate a computer-implemented process, thereby the instructions executed on the computer or other programmable apparatus provide steps for realizing the functions specified in one process or multiple processes and / or one block or multiple blocks. Figure 1 one process or multiple processes and / or Figure 1 blocks.

[0113] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative rather than restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the spirit and scope protected by the present invention and the claims. All of these fall within the protection scope of the present invention.

Claims

1. A method for extracting and fusing enhanced CT features of the liver based on vascular topology, characterized in that Including: Obtain enhanced CT images of the liver, where the enhanced CT images of the liver include enhanced CT images of the arterial phase, venous phase, delayed phase, and plain scan phase of the liver; Preprocess the enhanced CT images of the liver to obtain preprocessed enhanced CT images of the liver; Perform rigid registration on the preprocessed enhanced CT images of the liver to obtain registered enhanced CT images of the liver; Use the registered enhanced CT images of the liver as input, and perform time feature extraction based on a time feature extraction model to obtain a time feature vector; Segment and partition the registered enhanced CT images of the liver according to the vascular topological structure to obtain multiple liver partition images; Divide the time feature vector according to the multiple liver partition images to obtain partition feature vectors corresponding to each liver partition image; Use the partition feature vectors corresponding to each liver partition image as input, and perform spatial feature extraction and fusion based on a spatial feature extraction model to obtain a spatio-temporal feature vector; Obtain a classification result according to the spatio-temporal feature vector.

2. The liver enhanced CT feature extraction and fusion method based on vascular topology structure according to claim 1, characterized in that The enhanced CT images of the liver all use the same CT parameters, and the CT parameters include slice thickness and scanning voltage.

3. The liver enhanced CT feature extraction and fusion method based on vascular topology structure according to claim 1, wherein The preprocessing of the enhanced CT images of the liver to obtain preprocessed enhanced CT images of the liver includes: Perform Gaussian filtering on the enhanced CT images of the liver to obtain enhanced CT images of the liver after Gaussian filtering; Perform spatial resampling on the enhanced CT images of the liver after Gaussian filtering to unify them to a fixed resolution to obtain preprocessed enhanced CT images of the liver.

4. The method for extracting and fusing enhanced CT features of the liver based on vascular topology according to claim 1, characterized in that The rigid registration of the preprocessed enhanced CT images of the liver to obtain registered enhanced CT images of the liver includes: Use the enhanced CT image of the venous phase of the liver as the target image, and use the enhanced CT images of the arterial phase and delayed phase of the liver as the original images respectively for rigid registration. During the rigid registration process, use the gradient descent method to gradually optimize the objective function to make the similarity metric value between the original image and the target image the smallest to obtain the registered arterial phase image and the registered delayed phase image; The enhanced CT image of the venous phase of the liver, the registered arterial phase image, the registered delayed phase image, and the enhanced CT image of the plain scan phase of the liver form the registered enhanced CT image of the liver; The expression of the rigid registration is: ， Among them, 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: ， Among them, represents the similarity measure, represents the total number of pixel points in the image, represents the pixel value of the image before rigid registration at the position here, represents the pixel value of the image after rigid registration at the position here.

5. The method for extracting and fusing liver enhanced CT features based on vascular topology structure according to claim 1, wherein The partitioning of the registered enhanced CT images of the liver according to the vascular topological structure to obtain multiple liver partition images includes: Perform data cleaning on the registered enhanced CT images of the liver to obtain enhanced CT images of the liver after data cleaning; use the enhanced CT images of the liver after data cleaning as input, and automatically segment the liver and vascular regions based on a U-shaped network to obtain segmented enhanced CT images of the liver; Calculate the distance transformation matrix of the portal vein blood vessels according to the segmented enhanced CT images of the liver, and use the set of pixel points with the largest distance transformation in the distance transformation matrix as the vascular centerline; Traverse each pixel point on the vascular centerline. If a pixel point's neighborhood includes multiple branches, then determine it as a branch point, and all pixel points between two adjacent branch points form a blood vessel segment to form a vascular topological structure 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 assign it to the vascular segment with the closest distance to form multiple liver partition images.

6. The method for extracting and fusing liver enhanced CT features based on vascular topology structure according to claim 5, wherein The expression for calculating the distance transformation matrix of the portal vein blood vessel according to the segmented liver enhanced CT image is: ， Among them, represents the distance transformation matrix of the portal vein blood vessel, represents the pixel point coordinates, represents the blood vessel boundary coordinates, represents the blood vessel boundary set; The expression for the centerline of the blood vessel is: ， Among them, represents the blood vessel centerline; The expression for the vascular segment is: ， Among them, represents the k-th blood vessel segment, represents the i-th pixel point on the blood vessel centerline, represents the j-th branch point, represents the (j + 1)-th branch point; The expression for the liver partition image is: ， Among them, represents the liver partition image to which the pixel belongs, where represents the pixel coordinates, represents the coordinates of the pixel closest to the pixel on the i-th blood vessel segment at a distance, represents the i-th blood vessel segment, represents the set of blood vessel segments.

7. The method for extracting and fusing liver enhanced CT features based on vascular topology structure according to claim 1, wherein The partition feature vectors corresponding to each liver partition image obtained by dividing the temporal features according to multiple liver partition images are unified in vector length through average pooling.

8. The method for extracting and fusing liver enhanced CT features based on vascular topology structure according to claim 1, wherein The temporal feature extraction model uses a residual network, which includes an input layer, residual blocks, and an output layer connected in sequence; The input layer has 4 input channels, and each input channel corresponds to the venous phase, arterial phase, delayed phase, and plain scan phase in the registered liver enhanced CT image respectively; There are multiple residual blocks, and its processing expression is: ， Among them, represents the output feature, represents the input feature, 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, where the number of Transformer layers is 10.

9. The method for extracting and fusing liver enhanced CT features based on vascular topology structure according to claim 1, wherein Obtaining the classification result according to the spatio-temporal feature vector is realized through a fully connected network.

10. The liver enhanced CT feature extraction and fusion method based on vascular topology structure according to claim 1, characterized in that, It also includes pre-training the temporal feature extraction model, spatial feature extraction model, and fully connected network. The pre-training method includes: Obtain a liver enhanced CT image dataset; Label the liver, blood vessels, and tumor regions in the liver enhanced CT image dataset to obtain a labeled liver enhanced CT image dataset; Divide the labeled liver enhanced CT image dataset into a training set and a test set; Use the training set data as input to train the temporal feature extraction model, spatial feature extraction model, and fully connected network. During the training process, use the cross-entropy loss function to adjust the model parameters; Use the test set data as input to test the spatial feature extraction model and fully connected network, calculate the evaluation metrics according to the test results, and obtain the pre-trained temporal feature extraction model, spatial feature extraction model, and fully connected network based on the evaluation metrics; Among them, the evaluation metrics include accuracy, recall rate, and F1-score.

Images