A liver tumor segmentation method based on GTF convolution and multi-frequency dual attention

Abstract

The present invention discloses a liver tumor segmentation method based on GTF convolution and multi-frequency dual attention. This method aims to address the problems of liver tumors in terms of shape and size diversity and blurred boundaries, thereby achieving accurate liver tumor segmentation. The method comprises the following steps: S1: preprocessing abdominal CT data and dividing the CT data into a training set, a validation set, and a test set; S2: introducing GTF (Geometric Tight Framelet) and combining it with convolution operations to construct a GTF convolution; S3: fusing GTF convolution with an attention mechanism to construct a multi-frequency dual attention module; S4: fusing internal feature distillation loss and segmentation loss to construct a multi-scale distillation segmentation loss; S5: fusing GTF convolution, a multi-frequency dual attention module, and a multi-scale distillation segmentation loss to construct a GTF-UNet model; S6: using the training set, validation set, and test set to train, verify, optimize, and test the model. The present invention can effectively capture detailed features and multi-scale information, effectively improving the segmentation accuracy of liver tumors, and can be used for automatic segmentation of liver tumors in CT images.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and in particular to a liver tumor segmentation method based on GTF convolution and multi-frequency dual attention. Background Art

[0002] In the field of medical image processing, accurate segmentation of liver tumors is crucial for disease diagnosis, treatment planning, and prognosis assessment. With the rapid development of computer technology, image segmentation methods based on deep learning have become a research hotspot.

[0003] Traditional liver tumor segmentation methods often face numerous challenges. Liver tumors exhibit significant diversity in morphology, size, and location, and their boundaries are often blurred, making accurate definition of the tumor region extremely difficult. For example, some small tumors may have low contrast with surrounding liver tissue, while large tumors may have complex internal structures and irregular edges. Furthermore, abdominal CT images may be affected by noise, artifacts, and physiological differences between individuals, further reducing the accuracy and reliability of traditional segmentation methods.

[0004] In recent years, the U-Net architecture, based on convolutional neural networks (CNNs), has demonstrated promising performance in image segmentation tasks. However, U-Net and its derivatives still have limitations when it comes to liver tumor segmentation. Their ability to capture subtle features and multi-scale information needs improvement. For liver tumors with blurred boundaries, they struggle to precisely identify their edge contours, thus compromising segmentation accuracy. Furthermore, the network's relatively limited receptive field may not fully integrate global information within the image, resulting in poor segmentation results for tumors with unique morphologies or locations.

[0005] On the other hand, while traditional multiscale analysis methods such as wavelet transforms can extract information from different frequencies in images to a certain extent, they still lack the ability to integrate with deep learning models and adapt to the complex characteristics of liver tumors. To overcome these problems, it is necessary to explore new technologies and methods to improve the accuracy and robustness of liver tumor segmentation. Therefore, how to effectively improve the accuracy and reliability of liver tumor segmentation by leveraging the characteristics of wavelet transforms and the advantages of the U-Net architecture has become a key issue that needs to be addressed in the field of medical image segmentation. Summary of the Invention

[0006] In order to solve the above technical problems, the present invention provides a liver tumor segmentation method based on GTF convolution and multi-frequency dual attention, which can provide users with a more accurate, efficient and highly adaptable liver tumor segmentation solution.

[0007] Specifically, the method comprises the following steps:

[0008] S1: Preprocess abdominal CT data and divide the CT data into training set, validation set and test set;

[0009] S2: Introduce GTF (Geometric Tight Framelet) and combine it with convolution operation to construct GTF convolution; specifically, the following steps are included:

[0010] S2.1: Construct GTF transform and its inverse transform filter;

[0011] S2.2: Use GTF transform cascade decomposition and reconstruction to construct GTF convolution (GTFConv);

[0012] Preferably, S2.1 includes the following steps:

[0013] Construct the basic matrices W0, W1, and W2, which will serve as the key starting modules for generating a series of complex matrix structures. W0 contains the smoothness information of the image, W1 contains the direction and gradient information of the image, and W2 extracts the local changes and contrast information of the image that are opposite to the characteristics of W0.

[0014] definition Where i, j = 0, 1, 2, Represents the Kronecker product, which is combined to generate the GTF transform filter matrix group:

[0015] f0=W 0,0 ,

[0016] f 11 =W 2,1 , f 12 =W 1,2 ,

[0017]

[0018] The GTF transform filter bank is defined as in Is a low-pass filter is a high-pass filter, collectively referred to as f H ;Depend on

[0019] filters t filters=E

[0020] It can be seen that the GTF transform has orthogonality, and the GTF inverse transform filters are obtained t ;

[0021] Preferably, S2.2 includes the following steps:

[0022] Given a single-channel image X, a low-pass filter f based on GTF is L and high-pass filter f H , the GTF transform (GT) can be described as:

[0023] [X L ,X H ]=Conv([f L ,f H ],X)

[0024] The resolution of each frequency domain component is half of X, X L is the low-frequency component of X, X H is the high-frequency component of X; since the GTF transform is orthogonal, the inverse GTF transform (IGT) can be obtained by transposed convolution:

[0025] X=Conv-transposed([f L ,f H ],[X L ,X H ])

[0026] The cascaded GTF transform can be achieved by recursively decomposing the low-frequency components, where

[0027]

[0028] Then, the decomposed frequency domain components are convolved to obtain the intermediate variable Y, where w is the weight of the convolution kernel:

[0029]

[0030] GTF is a geometric tight frame wavelet. According to the linear properties of wavelet transform and inverse transform, we have:

[0031] GT(M+N)=GT(M)+GT(N)

[0032] IGT(M+N)=IGT(M)+IGT(N)

[0033] Use the linear properties of inverse wavelet transform to combine high-level output and low-level Y L :

[0034]

[0035] Finally, perform the inverse GTF transform to reconstruct the image and obtain the output Z of the GTF convolution:

[0036]

[0037] S3: Integrate GTF convolution and attention mechanism to build a multi-frequency dual attention module;

[0038] S4: Fusion of internal feature distillation loss and segmentation loss to construct multi-scale distillation segmentation loss;

[0039] S5: Fusing GTF convolution, multi-frequency dual attention module and multi-scale distillation segmentation loss to build the GTF-UNet model;

[0040] S6: Use the training set, validation set, and test set to train, validate, optimize, and test the model.

[0041] Preferably, S1 comprises the following steps:

[0042] S1.1: Preprocess the acquired CT data, including cropping to remove irrelevant areas, resampling to ensure consistent data size, using filtering techniques to remove noise from the data, and normalizing the data.

[0043] S1.2: Divide the preprocessed CT data into training set, validation set, and test set.

[0044] Preferably, S3 includes the following steps:

[0045] S3.1: Build a dual attention module;

[0046] S3.2: Fuse GTF convolution to build a multi-frequency dual attention module (MFDAM);

[0047] Preferably, S3.1 includes the following steps:

[0048] The dual attention module is composed of spatial attention and channel attention. The feature map first passes through the channel attention module to compress the spatial dimension and obtain the channel dimension information; then the output of the channel attention module is used as input to calculate the spatial attention to compress the channel dimension and obtain the spatial information of the target area; the dual attention module A cs The overall calculation process is as follows:

[0049] A c (F D )=σ(MLP(AvgPool(F D ))+MLP(MaxPool(F D )))

[0050]

[0051] A s (F D ′)=σ(f 7×7 ([AvgPool(FD ′); MaxPool(F D ′)]))

[0052]

[0053] That is F D =A cs (F D )

[0054] In the above formula, F D is the input feature, F D ′,F D ″ are the output of the channel attention module and the final output of the dual attention module, represents the element-by-element multiplication operation, σ represents the sigmoid activation function, f 7×7 represents the filter, A c , A s Denote the channel attention layer and the spatial attention layer respectively, A cs represents the dual attention module;

[0055] Preferably, S3.2 includes the following steps:

[0056] Based on the one-layer decomposition and reconstruction process of GTF convolution, dual attention modules are introduced at the original input feature level and the frequency domain sub-graph level to form a multi-frequency dual attention module. The specific implementation is as follows:

[0057] F′=A cs (Conv(F in ))

[0058] F L ,F H =GT(F in )

[0059] F Gout =IGT(A cs (Conv(F L ,F H )))

[0060]

[0061] Among them, F in , F out Represent the input features and output features respectively, F L , F H The input features are decomposed into high and low frequency sub-graphs by GTF transformation. The high and low frequency sub-graphs are subjected to convolution and dual attention modules to obtain attention weights, and then the GTF inverse transformation is performed to reconstruct F Gout , and finally concatenate it with the output F′ of the original input feature layer to get the final output Fout .

[0062] Preferably, S4 includes the following steps:

[0063] S4.1: Calculate internal feature distillation loss;

[0064] S4.2: Calculate segmentation loss;

[0065] S4.3: Construct multi-scale distillation segmentation loss;

[0066] Preferably, S4.1 includes the following steps:

[0067] The GTF-UNet main body consists of an encoding block, a decoding block, and a bottleneck. The block contains a convolutional layer. For the feature map in each block, the number of channels is evenly divided into two parts in the channel dimension. The first half is used as the shallow channel feature. The second half is used as the deep channel feature Internal feature distillation loss function L IFD Defined as:

[0068]

[0069] In the above formula, represents the feature map of the jth layer in the bth block, B represents the total number of blocks in the network, J represents the total number of layers within the block, and the calculation adopts the L2 norm; in this way, information distillation is performed from shallow channel features to deep channel features;

[0070] Preferably, S4.2 includes the following steps:

[0071] Segmentation loss integrates Dice and cross entropy mixed loss; segmentation loss L MSD The definition is as follows:

[0072]

[0073] L MSD =α×Dice(P,G)+(1-α)×CrossEntropy(P,G)

[0074] Where P represents the predicted segmentation map, G represents the true segmentation map, N represents the number of pixels, ∈ is a positive number, and α is the weight that balances the Dice loss and the cross entropy loss;

[0075] Preferably, S4.3 includes the following steps:

[0076] Using multi-scale supervision, integrating internal feature distillation loss function L IFD and segmentation loss L MSD , and obtain the multi-scale distillation segmentation loss L MDS As shown below:

[0077]

[0078] Among them, w j is the weight assigned to the jth scale, L MSD (j) is the segmentation loss at the j-th scale, and β is used to balance the segmentation loss and the internal feature distillation loss.

[0079] Preferably, S5 includes the following steps:

[0080] Taking the U-shaped network as the basic architecture, the GTFConv block is embedded in the encoder and decoder, and then MFDAM is fused at the bottleneck of the U-shaped network. Finally, the multi-scale distillation segmentation loss function is used for optimization learning to form the GTF-UNet model.

[0081] Preferably, S6 includes the following steps:

[0082] Use the training set and validation set to train and optimize the model to obtain the optimal model weight, and then use the test set to test the model.

[0083] The method of the present invention has the following beneficial effects: First, during the data preprocessing stage, by performing operations such as cropping, resampling, denoising, and normalizing the acquired abdominal CT images, data quality is improved, irrelevant information interference is reduced, and the data is rationally divided into training, validation, and test sets, laying a good foundation for subsequent model training and testing. During the model construction process, the GTF and convolution operations are innovatively integrated to construct the GTF convolution. On this basis, a multi-frequency dual-attention module is constructed based on the GTF convolution. Furthermore, the internal feature distillation loss and the segmentation loss are integrated to construct a multi-scale distillation segmentation loss. Finally, a GTF-UNet model is constructed based on the U-shaped network. These innovative designs enhance the model's performance in multiple dimensions, enabling it to demonstrate excellent capabilities in capturing detailed features and multi-scale information, significantly enhancing the accuracy of edge recognition, and effectively expanding the network's receptive field, enabling the network to accurately and efficiently focus on liver tumor regions. At the same time, feature redundancy is reduced, preventing invalid or duplicate features from dragging down model performance, significantly improving the accuracy of liver tumor segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the prior art and the drawings required for use in the embodiments. The following drawings are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0085] Figure 1This is a flowchart of a liver tumor segmentation method based on GTF convolution and multi-frequency dual attention of the present invention;

[0086] Figure 2 This is a model architecture diagram of a liver tumor segmentation method based on GTF convolution and multi-frequency dual attention in the present invention;

[0087] Figure 3 Schematic diagram of the GTF convolution structure of a liver tumor segmentation method based on GTF convolution and multi-frequency dual attention in the present invention;

[0088] Figure 4 Schematic diagram of the MFDAM structure of a liver tumor segmentation method based on GTF convolution and multi-frequency dual attention in the present invention; Specific implementation plan

[0089] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making creative work are within the scope of protection of the present invention.

[0090] The embodiment of the present invention provides a liver tumor segmentation method based on GTF convolution and multi-frequency dual attention, which is used to achieve accurate segmentation of liver tumors, assist clinical diagnosis and treatment decision-making, and improve the accuracy and efficiency of medical image processing.

[0091] Reference Figure 1 , the method comprises the following steps:

[0092] S1: Preprocess abdominal CT data and divide the CT data into training set, validation set and test set;

[0093] S2: Introduce GTF (Geometric Tight Framelet), combine it with convolution operation, and construct GTF convolution; further, refer to Figure 3 , S2 includes the following steps:

[0094] S2.1: Construct GTF transform and its inverse transform filter;

[0095] S2.2: Use GTF transform cascade decomposition and reconstruction to construct GTF convolution (GTFConv);

[0096] Furthermore, S2.1 includes the following steps:

[0097] Construct the basic matrices W0, W1, and W2, which will serve as the key starting modules for subsequently generating a series of complex matrix structures. W0 contains the smoothing information of the image, W1 contains the direction and gradient information of the image, and W2 extracts the local changes and contrast information of the image that are opposite to the characteristics of W0. Based on the characteristics of W0, W1, and W2, define the basic matrix generation function:

[0098]

[0099] The basic matrix generation function can be used to generate the following basic matrix:

[0100]

[0101] definition Where i, j = 0, 1, 2, Represents the Kronecker product. Here, the size of the basic matrix is ​​set to 3×3, thereby generating the GTF transformation filter matrix group:

[0102] f0=W 0,0 ,

[0103]

[0104] f 11 =W 2,1 , f 12 =W 1,2 ,

[0105]

[0106] The GTF transform filter bank is defined as in Is a low-pass filter is a high-pass filter, collectively referred to as f H ;Depend on

[0107] filters t filters=E

[0108] It can be seen that the GTF transform has orthogonality, and the GTF inverse transform filters are obtained t ;

[0109] Furthermore, S2.2 includes the following steps:

[0110] Given a single-channel image X, a low-pass filter f based on GTF is L and high-pass filter f H , setting the convolution step size to 2 and the padding to 1, the GTF transform (GT) can be described as:

[0111] [X L ,X H ]=Conv([f L ,f H ],X)

[0112] The output of the convolution has 18 channels, and the resolution of each channel is half of X. L is the low-frequency component of ω, X H is the high-frequency component of X; since the GTF transform is orthogonal, the inverse GTF transform (IGT) can be obtained by transposed convolution:

[0113] X=Conv-transposed([f L ,f H ],[X L ,X H ])

[0114] The cascaded GTF transform can be achieved by recursively decomposing the low-frequency components, where

[0115]

[0116] Then, a 5×5 convolution operation is performed on the decomposed frequency domain components to obtain the intermediate variable Y, where w is the weight of the 5×5 convolution:

[0117]

[0118] GTF is a geometric tight frame wavelet. According to the linear properties of wavelet transform and inverse transform, we have:

[0119] GT(M+N)=GT(M)+GT(N)

[0120] IGT(M+N)=IGT(M)+IGT(N)

[0121] Use the linear properties of inverse wavelet transform to combine high-level output and low-level Y L :

[0122]

[0123] Finally, perform the inverse GTF transform to reconstruct the image and obtain the output Z of the GTF convolution:

[0124]

[0125] The number of recursive decomposition layers of GTF convolution in the GTF-UNet model is 3, so the maximum value of i is 3. Set Z (4) =0;

[0126] S3: Integrate GTF convolution and attention mechanism to build a multi-frequency dual attention module;

[0127] S4: Fusion of internal feature distillation loss and segmentation loss to construct multi-scale distillation segmentation loss;

[0128] S5: Fusing GTF convolution, multi-frequency dual attention module and multi-scale distillation segmentation loss to build the GTF-UNet model;

[0129] S6: Use the training set, validation set, and test set to train, validate, optimize, and test the model.

[0130] Furthermore, S1 includes the following steps:

[0131] S1.1: Preprocess the acquired CT data, including cropping to remove irrelevant areas, resampling to ensure consistent data size, using filtering techniques to remove noise from the data, and normalizing the data.

[0132] Furthermore, the preprocessing of the collected CT data includes: cropping to remove irrelevant areas, resampling to ensure data size consistency; using filtering technology to denoise the data, and performing normalization operations on the data. This step specifically includes:

[0133] The collected data include public datasets and datasets provided by partner hospitals. Each case image completely covers the entire liver area. The CT data is cropped to remove irrelevant areas and resampled to 512×512 size. Then, Gaussian filtering is used to denoise the CT data, and the pixel values ​​of each data are normalized to a uniform range (0 to 1).

[0134] Set the mask corresponding to the liver and background area to 0, and the mask corresponding to the liver tumor area to 1;

[0135] Convert the labeled CT data stored in nii format into npz matrix format data;

[0136] S1.2: Divide the preprocessed CT data into training set, validation set, and test set;

[0137] Furthermore, the step of dividing the pre-processed CT data into a training set, a validation set, and a test set specifically includes:

[0138] The CT data were divided into a training set, a validation set, and a test set in a ratio of 8:1:1.

[0139] Further, refer to Figure 4 , S3 includes the following steps:

[0140] S3.1: Build a dual attention module;

[0141] S3.2: Fuse GTF convolution to build a multi-frequency dual attention module (MFDAM);

[0142] Furthermore, S3.1 includes the following steps:

[0143] The dual attention module is composed of spatial attention and channel attention. The feature map first passes through the channel attention module to compress the spatial dimension and obtain the channel dimension information; then the output of the channel attention module is used as input to perform spatial attention calculation to compress the channel dimension and obtain the spatial information of the target area. Overall calculation process A cs As shown below:

[0144] A c (F D )=σ(MLP(AvgPool(F D ))+MLP(MaxPool(F D )))

[0145]

[0146] A s (F D ′)=σ(f 7×7 ([AvgPool(F D ′); MaxPool(F D ′)]))

[0147]

[0148] That is F D =A cs (F D )

[0149] In the above formula, F D is the input feature, F D ′,F D ″ are the output of the channel attention module and the final output of the dual attention module, represents the element-by-element multiplication operation, σ represents the sigmoid activation function, f 7×7 A represents a convolution operation with a filter size of 7×7. c , A s Denote the channel attention layer and the spatial attention layer respectively, A cs represents the dual attention module;

[0150] Furthermore, S3.2 includes the following steps:

[0151] Based on the one-layer decomposition and reconstruction process of GTF convolution, dual attention modules are introduced at the original input feature level and the frequency domain sub-graph level to form a multi-frequency dual attention module. The specific implementation is as follows:

[0152] F′=A cs (Conv(F in ))

[0153] F L ,F H =GT(F in )

[0154] F Gout =IGT(A cs (Conv(F L ,F H )))

[0155]

[0156] Among them, F in , F out Represent the input features and output features respectively, F L , F H The input features are decomposed into high and low frequency sub-graphs by GTF transformation. The high and low frequency sub-graphs are subjected to convolution and dual attention modules to obtain attention weights, and then the GTF inverse transformation is performed to reconstruct F Gout , and finally concatenate it with the output F′ of the original input feature layer to get the final output F out .

[0157] Further, S4 includes the following steps:

[0158] S4.1: Calculate internal feature distillation loss;

[0159] S4.2: Calculate segmentation loss;

[0160] S4.3: Construct multi-scale distillation segmentation loss;

[0161] Furthermore, S4.1 includes the following steps:

[0162] The GTF-UNet main body consists of four encoding blocks, four decoding blocks and a bottleneck. Each block contains two convolutional layers. For the feature map in each block, the number of channels is evenly divided into two parts in the channel dimension. The first half is used as the shallow channel feature. The second half is used as the deep channel feature Internal feature distillation loss function L IFD Defined as:

[0163]

[0164] In the above formula, represents the feature map of the jth layer in the bth block, B represents the total number of blocks in the network, J represents the total number of layers within the block, and the calculation adopts the L2 norm; in this way, information distillation is performed from shallow channel features to deep channel features;

[0165] Furthermore, S4.2 includes the following steps:

[0166] Segmentation loss integrates Dice and cross entropy mixed loss; segmentation loss L MSD The definition is as follows:

[0167]

[0168] L MSD =α×Dice(P,G)+(1-α)×CrossEntropy(P,G)

[0169] Where P represents the predicted segmentation map, G represents the true segmentation map, N represents the number of pixels, ∈ is a positive number, and α is the weight that balances the Dice loss and the cross entropy loss;

[0170] Furthermore, S4.3 includes the following steps:

[0171] Using multi-scale supervision, integrating internal feature distillation loss function L IFD and segmentation loss L MSD , and obtain the multi-scale distillation segmentation loss L MDS As shown below:

[0172]

[0173] Among them, k is set to 4, w j is the weight assigned to the jth scale, L MSD (j) is the segmentation loss at the j-th scale, and β is used to balance the segmentation loss and the internal feature distillation loss.

[0174] Further, refer to Figure 2 , S5 includes the following steps:

[0175] Taking the U-shaped network as the basic architecture, the GTFConv block is embedded in the encoder and decoder, and then MFDAM is fused at the bottleneck of the U-shaped network. Finally, the multi-scale distillation segmentation loss function is used for optimization learning to form the GTF-UNet model.

[0176] Furthermore, S6 includes the following steps:

[0177] Use the training set and validation set to train and optimize the model to obtain the optimal model weights. Then import the optimal model weights into the test code and use the test set to test the model.

[0178] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make other equivalent modifications or substitutions without violating the spirit of the invention, and these equivalent modifications or substitutions are included in the scope defined by the claims.

Claims

1. A liver tumor segmentation method based on GTF convolution and multi-frequency dual attention, It is characterized by: The method comprises the following steps: S1: Preprocess abdominal CT data and divide the CT data into training set, validation set and test set; S2: Introduce GTF (Geometric Tight Framelet) and combine it with convolution operation to construct GTF convolution; specifically, the following steps are included: S2.1: Construct GTF transform and its inverse transform filter; S2.2: Use GTF transform cascade decomposition and reconstruction to construct GTF convolution (GTFConv); S2.1 includes the following steps: Construct the basic matrices W0, W1, and W2, which will serve as the key starting modules for generating a series of complex matrix structures. W0 contains the smoothness information of the image, W1 contains the direction and gradient information of the image, and W2 extracts the local changes and contrast information of the image that are opposite to the characteristics of W0. definition Where i, j = 0, 1, 2, Represents the Kronecker product, which is combined to generate the GTF transform filter matrix group: f0=W 0,0 , f 11 =W 2,1 ,f 12 =W 1,2 , The GTF transform filter bank is defined as in is the low-pass filter f L , is a high-pass filter, collectively referred to as f H ;Depend on filters t filters=E It can be seen that the GTF transform has orthogonality, and the GTF inverse transform filters are obtained t ; Among them, S2.2 includes the following steps: Given a single-channel image X, a low-pass filter f based on GTF is L and high-pass filter f H , the GTF transform (GT) can be described as: [X L ,X H ]=Conv([f L ,f H ],X) The resolution of each frequency domain component is half of X, X L is the low-frequency component of X, X H is the high-frequency component of X; since the GTF transform is orthogonal, the inverse GTF transform (IGT) can be obtained by transposed convolution: X=Conv-transposed([f L ,f H ],[X L ,X H ]) The cascaded GTF transform can be achieved by recursively decomposing the low-frequency components, where Then, the decomposed frequency domain components are convolved to obtain the intermediate variable Y, where w is the weight of the convolution kernel: GTF is a geometric tight frame wavelet. According to the linear properties of wavelet transform and inverse transform, we have: GT(M+N)=GT(M)+GT(N) IGT(M+N)=IGT(M)+IGT(N) Use the linear properties of inverse wavelet transform to combine high-level output and low-level Y L : Finally, perform the inverse GTF transform to reconstruct the image and obtain the output Z of the GTF convolution: S3: Integrate GTF convolution and attention mechanism to build a multi-frequency dual attention module; S4: Fusion of internal feature distillation loss and segmentation loss to construct multi-scale distillation segmentation loss; S5: Fusing GTF convolution, multi-frequency dual attention module and multi-scale distillation segmentation loss to build the GTF-UNet model; S6: Use the training set, validation set, and test set to train, validate, optimize, and test the model.

2. The liver tumor segmentation method based on GTF convolution and multi-frequency dual attention according to claim 1, characterized in that: S1 includes the following steps: S1.1: Preprocess the acquired CT data, including cropping to remove irrelevant areas, resampling to ensure consistent data size, using filtering techniques to remove noise from the data, and normalizing the data. S1.2: Divide the preprocessed CT data into training set, validation set, and test set.

3. The liver tumor segmentation method based on GTF convolution and multi-frequency dual attention according to claim 1, characterized in that: S3 includes the following steps: S3.1: Build a dual attention module; S3.2: Fuse GTF convolution to build a multi-frequency dual attention module (MFDAM); Among them, S3.1 includes the following steps: The dual attention module is composed of spatial attention and channel attention. The feature map first passes through the channel attention module to compress the spatial dimension and obtain the channel dimension information; then the output of the channel attention module is used as input to calculate the spatial attention to compress the channel dimension and obtain the spatial information of the target area; the dual attention module A cs The overall calculation process is as follows: A c (F D )=σ(MLP(AvgPool(F D ))+MLP(MaxPool(F D ))) A s (F D ′)=σ(f 7×7 ([AvgPool(F D ′);MaxPool(F D ′)])) That is F D =A cs (F D ) In the above formula, F D is the input feature, F D ′,F D '' are the output of the channel attention module and the final output of the dual attention module, respectively. represents the element-by-element multiplication operation, σ represents the sigmoid activation function, f 7×7 represents the filter, A c , A s Denote the channel attention layer and the spatial attention layer respectively, A cs represents the dual attention module; Among them, S3.2 includes the following steps: Based on the one-layer decomposition and reconstruction process of GTF convolution, dual attention modules are introduced at the original input feature level and the frequency domain sub-graph level to form a multi-frequency dual attention module. The specific implementation is as follows: F′=A cs (Conv(f in )) f L ,f H =GT(f in ) f Gout =IGT(A cs (Conv(F L ,F H ))) Among them, F in , F out Represent the input features and output features respectively, F L , F H The input features are decomposed into high and low frequency sub-graphs by GTF transformation. The high and low frequency sub-graphs are subjected to convolution and dual attention modules to obtain attention weights, and then the GTF inverse transformation is performed to reconstruct F Gout , and finally concatenate it with the output F′ of the original input feature layer to get the final output f out .

4. The liver tumor segmentation method based on GTF convolution and multi-frequency dual attention according to claim 1, characterized in that: S4 includes the following steps: S4.1: Calculate internal feature distillation loss; S4.2: Calculate segmentation loss; S4.3: Construct multi-scale distillation segmentation loss; Among them, S4.1 includes the following steps: The GTF-UNet main body consists of an encoding block, a decoding block, and a bottleneck. The block contains a convolutional layer. For the feature map in each block, the number of channels is evenly divided into two parts in the channel dimension. The first half is used as the shallow channel feature. The second half is used as the deep channel feature Internal feature distillation loss function L IFD Defined as: In the above formula, represents the feature map of the jth layer in the bth block, B represents the total number of blocks in the network, J represents the total number of layers within the block, and the calculation adopts the L2 norm; in this way, information distillation is performed from shallow channel features to deep channel features; Among them, S4.2 includes the following steps: Segmentation loss integrates Dice and cross entropy mixed loss; segmentation loss L MSD The definition is as follows: L MSD =α×Dice(P,G)+(1-α)×CrossEntropy(P,G) Where P represents the predicted segmentation map, G represents the true segmentation map, N represents the number of pixels, ∈ is a positive number, and α is the weight that balances the Dice loss and the cross entropy loss; Among them, S4.3 includes the following steps: Using multi-scale supervision, integrating internal feature distillation loss function L IFD and segmentation loss L MSD , and obtain the multi-scale distillation segmentation loss L MDS As shown below: Among them, w j is the weight assigned to the jth scale, L MSD (j) is the segmentation loss at the j-th scale, and β is used to balance the segmentation loss and the internal feature distillation loss.

5. The liver tumor segmentation method based on GTF convolution and multi-frequency dual attention according to claim 1, characterized in that: S5 includes the following steps: Taking the U-shaped network as the basic architecture, the GTFConv block is embedded in the encoder and decoder, and then MFDAM is fused at the bottleneck of the U-shaped network. Finally, the multi-scale distillation segmentation loss function is used for optimization learning to form the GTF-UNet model.

6. The liver tumor segmentation method based on GTF convolution and multi-frequency dual attention according to claim 1, characterized in that: S6 includes the following steps: Use the training set and validation set to train and optimize the model to obtain the optimal model weight, and then use the test set to test the model.

Images