3D liver vascular segmentation model and its establishment method based on enhanced attention mechanism context bridging

Abstract

A three-dimensional liver vessel segmentation model based on enhanced attention mechanism context bridging and its establishment method include: an input layer for receiving input three-dimensional liver medical CT image data to obtain image features; first to fourth image overlapping feature encoding layers for performing overlapping feature encoding operations on low-dimensional image features; first to fourth enhanced attention mechanism downsampling layers, which are alternately connected with the first to fourth image overlapping feature encoding layers in sequence, for feature extraction and generating four different-scale image features; first to fourth enhanced attention mechanism context bridging layers connected in series, for fusing multi-scale image features and further extracting features to obtain the association of local and global information of multi-scale image features; first to fourth image block expansion layers, which are used to re-divide high-dimensional image features into lower-dimensional image features while increasing the width, height, and depth of the image features; first to fourth enhanced attention mechanism upsampling layers, which are alternately connected with the first to fourth image block expansion layers in sequence, for recursively splicing and further extracting features from image features of different scales to obtain final image features; and an output layer, which is used to calculate the final image features and generate three-dimensional liver vessel segmentation results. The present invention can achieve accurate three-dimensional liver vessel segmentation and assist doctors in diagnosing diseases such as liver vessels and tumors.

Description

Technical Field

[0001] The present invention belongs to the field of computer vision technology, and specifically relates to a three-dimensional liver blood vessel segmentation model based on enhanced attention mechanism context bridging, an establishment method thereof, and an image segmentation method. Background Art

[0002] Over the past few decades, liver disease has become one of the leading causes of death worldwide due to increased life expectancy, sedentary lifestyles, and persistent overnutrition. Despite the development of vaccines and antiviral drugs, the global burden of liver disease continues to increase.

[0003] Because some severe liver diseases pose a significant threat to patients' lives, liver surgery is often required to remove the lesions. Early diagnosis and treatment are crucial. Before liver tumor resection, accurate localization of the hepatic vessels is crucial. Understanding the detailed location of the tumor within the vessels and the vessel diameter can determine the surgical outcome. Intraoperative vascular damage can cause rapid bleeding. If not controlled promptly, this can lead to short-term hemodynamic instability, increasing the need for intraoperative blood transfusions and even requiring a change in surgical approach. This not only increases the risk of transfusion-related complications but can also cause postoperative tumor recurrence and reduce survival rates. Therefore, accurate segmentation of liver vessels is a crucial prerequisite for developing surgical planning tools and medical visualization applications for tumor surgery. Accurate liver vessel segmentation provides a powerful guarantee for liver surgery, significantly improving patient survival rates and minimizing intraoperative complications.

[0004] However, due to the complex and highly interwoven vascular tree in the liver, accurately segmenting liver vessels has always been a challenging task. Manual liver vessel segmentation by doctors is a time-consuming and tedious task, which has significant limitations in clinical treatment where every second counts. Therefore, achieving accurate liver vessel segmentation by computer has always been a practical problem that needs to be solved urgently.

[0005] Traditional algorithms such as threshold segmentation, region growing, and image filtering are susceptible to interference from noise in medical images, making it difficult to achieve refined vascular segmentation. Deep neural network models offer automatic feature extraction, high-resolution spatial information processing, and sophisticated edge and shape detection capabilities, enabling more refined and accurate liver vascular segmentation.

[0006] Currently, liver vessel segmentation network models implemented domestically and internationally can be roughly categorized as CNN-based and Transformer-based. These two model structures exhibit distinct advantages and disadvantages in the task of liver vessel segmentation. CNN-based network architectures are computationally efficient, spatially invariant, and employ convolutional kernels for local perception, effectively processing details in liver vessel images. However, they struggle to capture dependencies between distant pixels in the image and insufficiently capture the overall orientation and distribution of vessels. Transformer-based network architectures utilize an attention mechanism to capture dependencies between all positions in the image, making them ideal for processing global information. This is particularly helpful for identifying and segmenting the complex structure of the entire liver's blood vessels. However, these architectures are computationally complex, require a large amount of data, and lack the ability to extract local features. Summary of the Invention

[0007] The present invention aims to overcome the above-mentioned shortcomings of the prior art and proposes a three-dimensional liver vascular segmentation model based on context bridging with enhanced attention mechanism, an establishment method thereof, an image segmentation method, and a segmentation device.

[0008] In the first aspect, the present invention proposes a 3D liver vessel segmentation model based on context bridging with enhanced attention mechanism, comprising:

[0009] The input layer is used to receive input three-dimensional liver medical CT image data to obtain image features;

[0010] The first to fourth image overlapping feature coding layers are used to perform overlapping feature coding operations on low-dimensional image features to obtain higher-dimensional image features while reducing the width, height, and depth of the image features;

[0011] The first to fourth enhanced attention mechanism downsampling layers are alternately connected with the first to fourth image overlapping feature encoding layers mentioned above to extract features and generate four different scale image features;

[0012] The first to fourth reinforced attention mechanism context bridge layers are connected in series to fuse multi-scale image features and further extract features to obtain the correlation between local and global information of multi-scale image features;

[0013] The first to fourth image block expansion layers are used to re-divide the high-dimensional image features into lower-dimensional image features while increasing the width, height, and depth of the image features;

[0014] The first to fourth enhanced attention mechanism upsampling layers are alternately connected with the first to fourth image block expansion layers mentioned above to recursively concatenate and further extract features from image features of different scales to obtain the final image features;

[0015] The output layer is used to calculate the final image features and generate 3D liver vessel segmentation results.

[0016] According to one embodiment of the present invention, the image overlapping feature encoding module layer includes a three-dimensional convolution operation and a layer normalization operation. The three-dimensional convolution operation refers to the convolution operation of the image features in the width, height, and depth dimensions to capture and utilize the three-dimensional spatial information in the data; the layer normalization operation refers to the normalization of each sample individually to stabilize the distribution of activation values, which helps to speed up the training process and improve model performance.

[0017] According to one embodiment of the present invention, the enhanced attention mechanism downsampling layer (or enhanced attention mechanism upsampling layer) does not itself change the image feature dimension. In the model workflow, it is combined with the image overlapping feature encoding layer (or image block expansion layer) to achieve the effect of feature extraction and simultaneous downsampling (or upsampling). The enhanced attention mechanism downsampling layer (or enhanced attention mechanism upsampling layer) specifically includes:

[0018] Layer standardization operation, which standardizes each sample separately;

[0019] An efficient self-attention module, based on the self-attention mechanism, uses a spatial reduction ratio to reduce the spatial resolution of input image features, thereby reducing the complexity of matrix multiplication and improving the computational efficiency of the self-attention mechanism;

[0020] The hybrid reinforced feedforward neural network embeds a depth-wise convolution (Depth-wise Conv) in the original feedforward neural network, connects three layers of layer normalization operations in series and adds three skip connections. The input is directly added to the output of each layer normalization operation to construct a residual module to ensure that the neural network has a positive effect.

[0021] According to one embodiment of the present invention, the enhanced attention mechanism context bridging layer specifically includes:

[0022] Image feature deformation and splicing, used to deform, unify and combine multi-scale image features;

[0023] Hybrid efficient attention feature extraction, used to extract features from mixed image features;

[0024] Residual connections are used to maintain the stability of gradients during model training and reduce the difficulty of training;

[0025] The hybrid reinforcement feedforward neural network is the same as the above hybrid reinforcement feedforward neural network and is used to improve the expressive power of the model.

[0026] In a second aspect, the present invention proposes a method for establishing a three-dimensional liver vessel segmentation model based on enhanced attention mechanism context bridging, which is applied to the three-dimensional liver vessel segmentation model based on enhanced attention mechanism context bridging. The method includes:

[0027] Acquire three-dimensional liver medical CT image data and corresponding liver vascular label data annotated by medical professionals;

[0028] Perform data augmentation on the acquired data and use them as training and validation sets for model training in a 4:1 distribution ratio.

[0029] Input the training set into the 3D liver vessel segmentation model based on reinforced attention mechanism context bridging for training. Simultaneously, the accuracy of the model obtained from each round of training is verified using the validation set. This process is iterated until the loss function reaches a predetermined target (or the model converges).

[0030] After the iteration is completed, the optimal three-dimensional liver vascular segmentation model is saved and output.

[0031] According to one embodiment of the present invention, the data enhancement operation includes the following:

[0032] Resampling, according to the set standard voxel spacing, bilinear interpolation operation is performed to generate image feature data that conforms to the specific physical interval.

[0033] Eigenvalue clipping is performed according to the specified valid eigenvalue range, limiting the data values ​​to the specified range to ensure that the data does not exceed the expected boundaries.

[0034] Space filling: fill the data that does not meet the requirements according to the specified width, height and depth to ensure that the input size is the same.

[0035] Random flipping randomly flips the width, height, and depth dimensions to ensure the randomness of the input image features and avoid model overfitting.

[0036] Random eigenvalue changes: random scaling and random offset operations are performed on image eigenvalues ​​to increase the diversity of training data and avoid model overfitting.

[0037] According to one embodiment of the present invention, the error setting requirements are as follows:

[0038] When judging whether the model training has converged, the model loss value on the training set is required to be no more than 0.1, the loss value on the validation set is required to be no more than 0.25, and the accuracy of the three-dimensional liver vascular segmentation results during the inference process is required to be greater than 70%.

[0039] According to one embodiment of the present invention, the loss function adopts the DiceCELoss loss function.

[0040] The DiceCELoss loss function is a loss function in the MONAI (Medical Open Network for AI) library, specifically designed for medical image segmentation tasks. It combines the Dice loss function and the cross-entropy loss function to address the limitations of using either loss function alone, thereby improving the performance of the segmentation model.

[0041] The Dice loss function measures the degree of overlap between the predicted segmentation results and the true segmentation labels, and its formula is:

[0042]

[0043] Cross entropy loss is used to measure the difference between the predicted probability distribution and the true probability distribution in classification tasks. Its formula is:

[0044]

[0045] in, Used to represent the predicted value, Used to represent real values.

[0046] Furthermore, the formula of the DiceCELoss loss function is:

[0047]

[0048] in, and It is a custom hyperparameter used to balance the contribution of the two loss functions.

[0049] In a third aspect, the present invention proposes a three-dimensional liver vessel segmentation method based on context bridging with enhanced attention mechanism, comprising the following steps:

[0050] Acquire three-dimensional liver medical CT image data to be segmented;

[0051] The 3D liver medical CT image data to be segmented is input into the trained 3D liver vascular segmentation network to obtain the vascular segmentation results.

[0052] Output a corresponding three-dimensional segmentation result based on the blood vessel segmentation result.

[0053] In a fourth aspect, the present invention proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned training method for a three-dimensional liver vascular segmentation model based on context bridging with enhanced attention mechanism.

[0054] In a fifth aspect, the present invention provides an electronic device, characterized in that it includes:

[0055] Memory, used to store application computer programs and the data generated by them;

[0056] A processor is used to execute the computer program stored in the memory so that the electronic device implements the above-mentioned training method of the three-dimensional liver blood vessel segmentation model based on enhanced attention mechanism context bridging.

[0057] The core of the present invention is to build a deep neural network based on context bridging of enhanced attention mechanism by scientifically combining CNN and Transformer model architectures. While ensuring excellent recognition and segmentation capabilities of the complex structure of liver blood vessels, it reduces computational complexity, ensures spatial invariance, fully extracts detailed features of liver blood vessels, and performs specific optimization on the feature extraction module. After a certain amount of training, the present invention can achieve accurate segmentation of three-dimensional liver blood vessels, assist doctors in diagnosing diseases such as liver blood vessels and tumors, and lay a solid foundation for the application of liver surgery planning and medical visualization. Furthermore, the present invention can improve the diagnostic efficiency of doctors, reduce surgical risks, and protect the life, health and safety of patients.

[0058] Beneficial effects of the present invention:

[0059] This invention achieves multi-scale feature fusion and efficient computing and memory usage by scientifically combining CNN and Transformer. It has powerful global and local feature extraction capabilities, the ability to handle long-range dependencies, and excellent robustness and generalization capabilities. It helps doctors achieve more accurate and detailed three-dimensional liver vascular segmentation and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 Schematic diagram of the structure of the three-dimensional liver blood vessel segmentation model based on enhanced attention mechanism context bridging of the present invention;

[0061] Figure 2 Schematic diagram of a training method for a three-dimensional liver vessel segmentation model based on enhanced attention mechanism context bridging of the present invention;

[0062] Figure 3 Schematic diagram of the steps of a training method for a 3D liver vessel segmentation model based on enhanced attention mechanism context bridging in one embodiment of the present invention;

[0063] Figure 4 Schematic diagram of the steps of a training method for a 3D liver vessel segmentation model based on enhanced attention mechanism context bridging in one embodiment of the present invention;

[0064] Figure 5 FIG2 is a schematic diagram of the steps of a three-dimensional liver vessel segmentation method based on enhanced attention mechanism context bridging in one embodiment of the present invention. DETAILED DESCRIPTION

[0065] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0066] Example 1

[0067] This embodiment proposes a 3D liver vessel segmentation model based on context bridging with enhanced attention mechanism. Figure 1 As shown, in one embodiment of the invention, a 3D liver vessel segmentation model 10 based on enhanced attention mechanism context bridging specifically includes: an input layer 11, which is used to receive input 3D liver medical CT image data to obtain image features; an image overlapping feature encoding layer 12, which is used to perform overlapping feature encoding operations on low-dimensional image features to obtain higher-dimensional image features while reducing the width, height, and depth of the image features; an enhanced attention mechanism downsampling layer 13, which is alternately connected to the image overlapping feature encoding layer 12 in sequence, and is used to extract features and generate image features of four different scales; an enhanced attention mechanism context bridging layer 14, which is used to fuse multi-scale image features and further extract features to obtain the association between local and global information of multi-scale image features; an image block expansion layer 15, which is used to re-divide the high-dimensional image features into lower-dimensional image features while increasing the width, height, and depth of the image features; an enhanced attention mechanism upsampling layer 16, which is alternately connected to the image block expansion layer 15 in sequence, and is used to recursively splice and further extract features of image features of different scales to obtain final image features; and an output layer 17, which is used to calculate the final image features and generate a 3D liver vessel segmentation result.

[0068] Example 2

[0069] For example, see Figure 2 As shown, this embodiment discloses a method for establishing a 3D liver vessel segmentation model based on enhanced attention mechanism context bridging, which is applied to training a 3D liver vessel segmentation model based on enhanced attention mechanism context bridging, and specifically includes the following steps:

[0070] Step S201, obtaining liver three-dimensional medical CT image data and corresponding liver blood vessel label data annotated by medical professionals;

[0071] Step S202: perform data augmentation on the acquired data and use them as the training set and validation set for model training in a 4:1 distribution ratio.

[0072] Step S203: Input the training set into the enhanced attention mechanism context bridging-based 3D liver vessel segmentation model for training. Simultaneously, the accuracy of the model obtained from each round of training is verified using the validation set. This process is iterated until the loss function reaches a predetermined target (or the model converges).

[0073] Step S204: After the iteration is completed, the optimal three-dimensional liver blood vessel segmentation model is saved and output.

[0074] Specifically, in one embodiment of the invention, Figure 3 As shown in FIG, the method for establishing a 3D liver vascular segmentation model based on context bridging with enhanced attention mechanism includes the following steps:

[0075] Step S301, performing a three-dimensional convolution operation on the image features to obtain first processed data;

[0076] Step S302, performing a layer normalization operation on the first processed data to obtain second processed data;

[0077] Step S303, performing an efficient self-attention operation on the second processed data to obtain third processed data;

[0078] Step S304: The third processed data is processed through a hybrid reinforcement neural network to obtain fourth processed data;

[0079] Determine whether the fourth processed data reaches the predetermined dimension. If not, execute step S305, save the intermediate results and continue to loop to step S301; if it reaches the predetermined dimension, execute step S306, splice all intermediate results as the fifth processed data and output it.

[0080] Specifically, in one embodiment of the invention, Figure 4 As shown, the method for establishing a 3D liver vascular segmentation model based on context bridging with enhanced attention mechanism also includes the following steps:

[0081] Step S401, deforming and splicing the input multi-scale image features to form first processed data;

[0082] Step S402, performing a hybrid high-efficiency attention operation on the first processed data to obtain second processed data;

[0083] Step S403: The second processed data is processed through a hybrid reinforcement feedforward neural network to obtain third processed data;

[0084] Step S404: The third processed data is added to the first processed data to form fourth processed data.

[0085] Example 3

[0086] This embodiment provides a three-dimensional liver vessel segmentation method based on enhanced attention mechanism context bridging, which uses the liver vessel segmentation model trained by the training method of the three-dimensional liver vessel segmentation model based on enhanced attention mechanism context bridging described in Examples 1 and 2, wherein, Figure 5 As shown, the segmentation method includes the following steps:

[0087] Step S501, obtaining three-dimensional liver medical CT image data to be segmented;

[0088] Step S502: inputting the three-dimensional liver medical CT image data to be segmented into a trained three-dimensional liver blood vessel segmentation network to obtain a blood vessel segmentation result;

[0089] Step S503: outputting a corresponding three-dimensional segmentation result based on the blood vessel segmentation result.

[0090] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of such features. "Multiple" means two or more, unless otherwise specifically defined.

[0091] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. Throughout this specification, the schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, those skilled in the art may combine and integrate different embodiments or examples described in this specification, as well as features from different embodiments or examples, without conflicting interpretations. Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. Throughout this specification, the schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, those skilled in the art may combine and integrate different embodiments or examples and features of different embodiments or examples described in this specification without mutual contradiction.

[0092] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0093] The logic and / or steps represented in a flowchart or otherwise described herein, for example, can be considered a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" is any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (not exhaustive) of computer-readable media include: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and a portable compact disc read-only memory (CDROM). Furthermore, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing it in another suitable manner if necessary, and then storing it in a computer memory.

[0094] It should be understood that various components of the present invention may be implemented using hardware, software, firmware, or a combination thereof. In the aforementioned embodiments, multiple steps or methods may be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any of the following technologies known in the art, or a combination thereof, may be used: a discrete logic circuit having logic gates for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gates, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0095] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0096] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or two or more units may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or in the form of software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.

[0097] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and are not to be construed as limiting the present invention. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

[0098] Example 4

[0099] This embodiment proposes a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the three-dimensional liver blood vessel segmentation method based on enhanced attention mechanism context bridging of embodiment 3 is implemented.

[0100] Example 5

[0101] This embodiment provides an electronic device, including:

[0102] Memory, used to store application computer programs and the data generated by them;

[0103] A processor is used to execute the computer program stored in the memory so that the electronic device implements the above-mentioned training method of the three-dimensional liver blood vessel segmentation model based on enhanced attention mechanism context bridging.

[0104] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A system for 3D liver vascular segmentation model based on context bridging with enhanced attention mechanism, characterized by: include: The input layer is used to receive input three-dimensional liver medical CT image data to obtain image features; The first to fourth image overlapping feature coding layers are used to perform overlapping feature coding operations on low-dimensional image features to obtain higher-dimensional image features while reducing the width, height, and depth of the image features; The first to fourth enhanced attention mechanism downsampling layers are alternately connected with the first to fourth image overlapping feature encoding layers mentioned above to extract features and generate four different scale image features; The first to fourth reinforced attention mechanism context bridge layers are connected in series to fuse multi-scale image features and further extract features to obtain the correlation between local and global information of multi-scale image features; The first to fourth image block expansion layers are used to re-divide the high-dimensional image features into lower-dimensional image features while increasing the width, height, and depth of the image features; The first to fourth enhanced attention mechanism upsampling layers are alternately connected with the first to fourth image block expansion layers mentioned above to recursively concatenate and further extract features from image features of different scales to obtain the final image features; The output layer is used to calculate the final image features and generate three-dimensional liver vessel segmentation results; The enhanced attention mechanism downsampling layer or the enhanced attention mechanism upsampling layer does not change the image feature dimension itself. In the model workflow, it is combined with the image overlapping feature encoding layer or the image block expansion layer to achieve the effect of feature extraction and simultaneous downsampling or upsampling. Specifically, it includes: Layer standardization operation, which standardizes each sample separately; An efficient self-attention module uses the spatial reduction ratio based on the self-attention mechanism to reduce the spatial resolution of input image features, thereby reducing the complexity of matrix multiplication and improving the computational efficiency of the self-attention mechanism; The hybrid reinforced feedforward neural network embeds a deep convolution in the original feedforward neural network, connects three layers of layer normalization operations in series, and adds three skip connections. The input is directly added to the output of each layer normalization operation to construct a residual module to ensure that the neural network has a positive effect. Strengthen the attention mechanism context bridge layer, including: Image feature deformation and splicing, used to deform, unify and combine multi-scale image features; Hybrid efficient attention feature extraction, used to extract features from mixed image features; Residual connections are used to maintain the stability of gradients during model training and reduce the difficulty of training; A hybrid reinforced feedforward neural network is used to improve the expressive power of the model.

2. The method for establishing a system for a 3D liver vascular segmentation model based on enhanced attention mechanism context bridging according to claim 1, characterized in that: The steps include: S1. Obtaining three-dimensional liver medical CT image data and its corresponding liver vascular label data annotated by medical professionals; S2. Perform data augmentation on the acquired data and use them as the training set and validation set for model training in a 4:1 distribution ratio. S3. Inputting the training set into the 3D liver vascular segmentation model based on enhanced attention mechanism context bridging for training, and simultaneously using the validation set to verify the accuracy of the model obtained from each round of training; iterating this process until the loss function reaches a predetermined target, which is called model convergence; S4. After the iteration is completed, save and output the optimal 3D liver vascular segmentation model.

3. The method for establishing a system for a 3D liver vascular segmentation model based on enhanced attention mechanism context bridging according to claim 2, characterized in that: The data enhancement operations include resampling, eigenvalue clipping, space filling, random flipping, and random eigenvalue changes.

4. The method according to claim 2, characterized in that The loss function adopts the DiceCELoss loss function. The Dice loss function measures the degree of overlap between the predicted segmentation result and the true segmentation label. Its formula is: Cross entropy loss is used to measure the difference between the predicted probability distribution and the true probability distribution in classification tasks. Its formula is: in, represents the index of the image pixel, Used to represent the predicted value, Used to represent real values; The formula of the DiceCELoss loss function is: in, and It is a custom hyperparameter used to balance the contribution of the two loss functions.

5. The method according to claim 2, characterized in that The condition for judging whether the model training has converged is that the loss value of the model on the training set does not exceed 0.1, and the loss value on the validation set is not greater than 0.

25.

6. A method for 3D liver vessel segmentation based on enhanced attention mechanism context bridging, comprising the following steps: Acquire three-dimensional liver medical CT image data to be segmented; Inputting the three-dimensional liver medical CT image data to be segmented into the trained three-dimensional liver vascular segmentation network to obtain the vascular segmentation result; Output a corresponding three-dimensional segmentation result based on the blood vessel segmentation result.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the three-dimensional liver blood vessel segmentation method based on enhanced attention mechanism context bridging as described in claim 6 is implemented.

8. An electronic device, characterized in that: include: Memory, used to store application computer programs and the data generated by them; A processor is configured to execute the computer program stored in the memory so that the electronic device implements the three-dimensional liver vessel segmentation method based on enhanced attention mechanism context bridging as claimed in claim 6.

Images