A method and system for constructing a brain tumor image segmentation model

Abstract

The present invention provides a method and system for constructing a brain tumor image segmentation model. The method and system are to construct an initial brain tumor image segmentation model including an encoder and a decoder, collect brain tumor head images and perform preprocessing, and divide them into a training set and a validation set; input the initial brain tumor image segmentation model according to a preset loss and the training set to complete the training of the model; and verify that the model meets the preset conditions through the validation set to obtain the optimal brain tumor image segmentation model. Compared with the prior art, the present invention combines the neural network technology with the special design of the encoder and decoder in the initial brain tumor image segmentation model, so that during the training process, the location characteristics of the brain tumor image can be effectively identified, the model parameters are reduced, and the efficiency and accuracy of image segmentation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method and system for constructing a brain tumor image segmentation model. Background Art

[0002] Brain tumor is a disease that seriously threatens the health of patients. In recent years, early brain tumors often have no obvious symptoms, which makes it difficult to detect and diagnose them in time. Brain tumor detection refers to the accurate positioning and segmentation of brain tumor areas by artificial or other means. Therefore, establishing an efficient and accurate brain tumor detection model is of great significance to improving the early detection rate and treatment effect of brain tumors. In recent years, deep learning has made great progress in the field of signal processing. The good performance of deep learning in the field of computer vision has enabled it to be widely used in the field of medical imaging, providing new opportunities for the classification and segmentation of brain tumors. In order to accurately detect brain tumors, deep learning can learn a large amount of medical images and clinical data to obtain tiny feature characteristics.

[0003] Some scholars have developed a brain tumor segmentation model based on deep learning to analyze the patient's brain images and achieve efficient segmentation and classification of brain tumors. The model will be trained using medical image datasets, and the accuracy and effectiveness of the model will be improved by optimizing the network structure and parameters. One important aspect is to collect a large amount of brain imaging data, including different types of MRI, CT, and PET images. These data will contain normal brain images of patients with pathological features so that the model can learn different types of lesions. Convolutional neural networks (CNNs) are often used as the core architecture of deep brain tumor segmentation models. CNNs can effectively process medical images and extract important features in images through convolution and pooling operations to achieve the localization of brain tumors.

[0004] At present, it is difficult to efficiently segment brain tumors due to factors such as uneven distribution of tumor structure information and location, complex boundaries, and variable lesions. Generally speaking, the correlation between different modality features and the correlation between global features and local features have not been fully considered.

[0005] Therefore, providing a method and system for constructing a brain tumor image segmentation model to solve the above problems is an urgent problem to be solved by those skilled in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a method for constructing a brain tumor image segmentation model. The method has clear logic, is safe, effective, reliable and easy to operate, takes into account the correlation between different modality features, and effectively improves the efficiency and accuracy of image segmentation.

[0007] Based on the above objectives, the technical solution provided by the present invention is as follows:

[0008] A method for constructing a brain tumor image segmentation model comprises the following steps:

[0009] Build an initial brain tumor image segmentation model including an encoder and a decoder;

[0010] The preprocessed brain tumor images were divided into training set and validation set;

[0011] Complete model training according to the training set, the initial brain tumor image segmentation model and the preset loss;

[0012] A brain tumor image segmentation model that meets preset conditions is determined based on the verification set.

[0013] Preferably, the initial brain tumor image segmentation model further comprises: a preprocessing convolution module;

[0014] The preprocessing convolution module is used to stretch the channel length of brain tumor images of different modes;

[0015] The encoder is used to compress the size of the brain tumor image to extract features between space and channels, and output the result;

[0016] The decoder is used to segment the features of the brain tumor image according to the output result of the encoder.

[0017] Preferably, the encoder comprises: a plurality of efficient convolution modules;

[0018] The efficient convolution module includes: a first point-by-point convolution layer, a channel-by-channel convolution layer, and a second point-by-point convolution layer;

[0019] The channel-by-channel convolution layer is arranged between the first point-by-point convolution layer and the second point-by-point convolution layer;

[0020] Each convolutional layer applies an activation function with group normalization.

[0021] Preferably, the encoder further comprises: an attention module;

[0022] The attention module is arranged between the channel-by-channel convolution layer and the second point-by-point convolution layer;

[0023] The attention module is used to apply channel-space attention to the maximum representation in the channel-space field.

[0024] Preferably, the decoder comprises: a plurality of up-convolution modules and a Conv3D module;

[0025] The upper convolution module is connected to the efficient convolution module and the Conv3D module respectively;

[0026] The up-convolution module cooperates with the efficient convolution module to reconstruct the position features in the brain tumor image;

[0027] The Conv3D module is used to obtain the confidence of different position features in the brain tumor image area according to the position features in the brain tumor image to segment the brain tumor image.

[0028] Preferably, the up-convolution module comprises: a transposed three-dimensional convolution layer, a first three-dimensional convolution layer, and a second three-dimensional convolution layer;

[0029] The transposed three-dimensional convolution layer is used to concatenate the output result with the output result of the efficient convolution module as input data;

[0030] The first three-dimensional convolution layer and the second three-dimensional convolution layer are used to reconstruct the position features in the brain tumor image according to the input data output;

[0031] Each convolutional layer applies an activation function with group normalization.

[0032] Preferably, the model training is completed according to the training set, the initial brain tumor image segmentation model and the preset loss, specifically:

[0033] The initial brain tumor image segmentation model is trained according to the cross entropy loss, the DICE loss function, the preset trade-off factor and the training set to minimize the total loss.

[0034] Preferably, the model training formula is specifically:

[0035] ;

[0036] in, is the total loss, is the cross entropy loss between the brain tumor labels segmented by the model and the true brain tumor labels, is the DICE loss between the two, is a trade-off factor.

[0037] A brain tumor image segmentation model construction system, comprising:

[0038] A model building module, used to build an initial brain tumor image segmentation model including an encoder and a decoder;

[0039] A sample image module is used to divide the preprocessed brain tumor images into a training set and a validation set;

[0040] A model training module, used to complete model training according to the training set, the initial brain tumor image segmentation model and a preset loss;

[0041] A verification module is used to determine a brain tumor image segmentation model that meets preset conditions based on the verification set.

[0042] The method for constructing a brain tumor image segmentation model provided by the present invention is to construct an initial brain tumor image segmentation model including an encoder and a decoder, collect brain tumor images and preprocess them, and divide them into a training set and a validation set; input the training set into the initial brain tumor image segmentation model according to a preset loss and to complete the training of the model; and verify that the model meets the preset conditions through the validation set to obtain the optimal brain tumor image segmentation model.

[0043] Compared with the prior art, the present invention combines neural network technology with the special design of the encoder and decoder in the initial brain tumor image segmentation model, so that during the training process, the location characteristics of the brain tumor image can be effectively identified, the difficulty of parameter acquisition is reduced, and the efficiency and accuracy of image segmentation are improved.

[0044] The present invention also provides a brain tumor image segmentation model construction system, which has the same technical concept as the method, solves the same technical problem, and should have the same beneficial effects, so it will not be described in detail here. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0046] Figure 1 A flowchart of a method for constructing a brain tumor image segmentation model provided by an embodiment of the present invention;

[0047] Figure 2 A schematic diagram of the overall structure of a brain tumor image segmentation model provided by an embodiment of the present invention;

[0048] Figure 3 A schematic diagram of the structure of an encoder provided by an embodiment of the present invention;

[0049] Figure 4 A schematic diagram of the structure of a decoder provided by an embodiment of the present invention;

[0050] Figure 5 A schematic diagram of the structure of a brain tumor image segmentation model building system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] The embodiments of the present invention are written in a progressive manner.

[0053] The embodiment of the present invention provides a method and system for constructing a brain tumor image segmentation model, which mainly solves the technical problem in the prior art that the correlation between different modality features and the correlation between global features and local features are insufficient, resulting in low efficiency and accuracy of brain tumor image segmentation.

[0054] like Figure 1 As shown, a method for constructing a brain tumor image segmentation model includes the following steps:

[0055] S1. Construct an initial brain tumor image segmentation model including an encoder and a decoder;

[0056] S2. The preprocessed brain tumor images are divided into a training set and a validation set;

[0057] S3. Complete model training according to the training set, the initial brain tumor image segmentation model and the preset loss;

[0058] S4. Determine a brain tumor image segmentation model that meets preset conditions based on the validation set.

[0059] In step S1 to step S4, an initial brain tumor image segmentation model is constructed, and the initial brain tumor image segmentation model includes an encoder and a decoder; a brain tumor image is obtained through a sample image module, or a brain tumor image is called in a medical system, specifically a brain tumor image obtained by CT or magnetic resonance examination; the brain tumor image is preprocessed to eliminate irrelevant information in the image, restore useful real information, enhance the detectability of relevant information and simplify data to the maximum extent; the preprocessed brain tumor image is then divided into a training set and a validation set; the initial brain tumor image segmentation model is trained through the training set and a preset loss; and the brain tumor image segmentation model that meets the preset conditions (with the minimum loss) is determined as the optimal model through the validation set.

[0060] like Figure 2 As shown, preferably, the initial brain tumor image segmentation model further includes: a preprocessing convolution module;

[0061] A preprocessing convolution module to stretch the channel length of brain tumor images of different modes;

[0062] An encoder, used for compressing the size of the brain tumor image to extract features between spaces and channels, and outputting the result;

[0063] The decoder is used to segment the features of the brain tumor image based on the output of the encoder.

[0064] During actual application, a preprocessing convolution module is also set up in the initial brain tumor image segmentation model. The channel length of magnetic resonance brain tumor images of different modes is stretched through the preprocessing convolution module; the magnetic resonance brain tumor image is compressed through the encoder to extract the features between space and channels, thereby reducing the number of parameters while maintaining the perception accuracy; the decoder combines the output results of the encoder in the channel dimension to reconstruct the positional features of the tumor image.

[0065] like Figure 3 As shown, preferably, the encoder includes: a plurality of efficient convolution modules;

[0066] The efficient convolution module includes: a first point-by-point convolution layer, a channel-by-channel convolution layer, and a second point-by-point convolution layer;

[0067] The channel-by-channel convolution layer is arranged between the first point-by-point convolution layer and the second point-by-point convolution layer;

[0068] Each convolutional layer applies an activation function with group normalization.

[0069] In practice, the EB module consists of three convolutional layers: First, a 3D point-wise convolutional layer with a stride of 1, an output channel size of "expand_c", and a kernel size of 1. Second, a 3D channel-wise convolution with a stride of 2, a kernel size of 1, and feature channels remain unchanged. Finally, a 3D point-wise convolution with an output channel number of "out_channel" is used to generate features. Each convolutional layer applies ReLU6 and group normalization (GN), with 2 groups.

[0070] In this embodiment, the four efficient convolution modules (EB) based on 3D depthwise separable convolution have 32, 64, 96 and 128 output channels respectively, and they also have 64, 128, 256 and 512 expansion scales (expand_c) to expand to higher dimensional feature space.

[0071] Preferably, the encoder further comprises: an attention module;

[0072] The attention module is set between the channel-wise convolution layer and the second point-wise convolution layer;

[0073] An attention module is used to apply channel-space attention to the maximum representation in the channel-space field.

[0074] In actual application, between the channel-by-channel convolution layer and the second point-by-point convolution layer, the EB module establishes an attention module through Squeeze-and-Excitation (SE). The Squeeze-and-Excitation (SE) module is used to apply channel-space attention to the maximum representation. It consists of 3D adaptive average pooling and two 3D convolution layers with a stride of 1. The SE module is placed after the channel-by-channel convolution layer to achieve the maximum representation in the channel-space field.

[0075] In this embodiment, a skip connection is added to the low-level features of each EB module to extract more imaging details. The skip connection (SKIP) module consists of a depthwise 3D convolution with a stride of 2 and a selectable 3D convolution layer with a stride of 1.

[0076] like Figure 4 As shown, preferably, the decoder includes: a plurality of up-convolution modules and a Conv3D module;

[0077] The upper convolution module is connected to the efficient convolution module and the Conv3D module respectively;

[0078] The up-convolution module cooperates with the efficient convolution module to reconstruct the location features in the brain tumor image;

[0079] The Conv3D module is used to obtain the confidence of different position features of the brain tumor image region according to the position features in the brain tumor image to segment the brain tumor image.

[0080] In practice, the decoder consists of four upconvolution (UP) modules with 32, 64, 96, and 128 channels, and one Conv3D module. The upconvolution module includes a transposed 3D convolution with a stride of 2 and a kernel size of 2, and two 3D convolutional layers followed by group normalization (GN) and ReLU. The kernel size of the 3D convolutional layer is 3 and the stride is 1.

[0081] Preferably, the up-convolution module comprises: a transposed three-dimensional convolution layer, a first three-dimensional convolution layer, and a second three-dimensional convolution layer;

[0082] The transposed 3D convolution layer is used to concatenate the output result with the output result of the efficient convolution module as input data;

[0083] A first three-dimensional convolutional layer and a second three-dimensional convolutional layer are used to output position features in the reconstructed brain tumor image according to the input data;

[0084] Each convolutional layer applies an activation function with group normalization.

[0085] In actual application, the UP module will reconstruct the location characteristics of the tumor by splicing the two outputs of the corresponding EB module and the previous module as input. The last UP module will restore the feature size to 160×160×128 and reduce the output channel to 3 through the final Conv3D module, which represents the confidence of different voxels in the sub-region of the tumor.

[0086] In this embodiment, the splicing method is specifically as follows:

[0087] Input to the corresponding EB module and i indivual UP module input Perform a union operation along the channel dimension and use the union result as the input of the Conv3D module , as shown in formula 1.

[0088] .

[0089] Preferably, model training is completed according to the training set, the initial brain tumor image segmentation model and the preset loss, specifically:

[0090] The initial brain tumor image segmentation model is trained according to the cross entropy loss, DICE loss function, preset trade-off factors and training sets to minimize the total loss.

[0091] Preferably, the model training formula is specifically:

[0092] ;

[0093] in, is the total loss, is the cross entropy loss between the brain tumor labels segmented by the model and the true brain tumor labels, is the DICE loss between the two, is a trade-off factor.

[0094] In actual application, in the brain tumor segmentation task, the class imbalance problem is one of the key issues due to the small key area and complex structure. The above loss is used to train the initial brain tumor segmentation model and minimize it during the training process.

[0095] The DICE loss function is a common solution to the problem of inter-class imbalance, which allows the network to focus on sample classes with smaller areas. is the trade-off factor. During training, we set this factor to , and uses automatic mixed precision training methods to accelerate model training.

[0096] It should be noted that the first algorithm is used in the model training process, and the second algorithm is used in the model testing process. In both the training and testing processes, the center cropping method is used to resize the MRI image to 160×160×128 to remove the extra black parts around the brain tumor. In the model testing process, we use the full MRI image of 240×240×155 as input.

[0097] In this embodiment, the model training process is as follows:

[0098] enter:

[0099] MRI scans from the training dataset: .

[0100] MRI image scans of the validation dataset: .

[0101] The corresponding real sample labels .

[0102] Efficient Brain Tumor Segmentation Model: EUN .

[0103] Initial model parameters:

[0104] Optimizer

[0105] Output:

[0106] Segmentation results:

[0107] The trained model parameters are: ;

[0108] The pseudo code for model training is as follows:

[0109] for i=1, 2, ..., N do

[0110] for n=1, 2, ..., M do

[0111] S X =EUN ( X )

[0112] loss=L ( S x , GT )

[0113]

[0114] end

[0115] for n=1, 2, ..., I do

[0116] S Y =EUN ( Y )

[0117] loss=L ( S y , GT )

[0118] end

[0119] end

[0120] return .

[0121] In this embodiment, the model verification process is as follows:

[0122] enter:

[0123] MRI scans of the test dataset: .

[0124] The corresponding real sample labels .

[0125] Trained efficient brain tumor segmentation model: EUN .

[0126] The trained model parameters are:

[0127] Output:

[0128] Segmentation results:

[0129] The pseudo code for model validation is as follows:

[0130] for i=1, 2, ..., N do

[0131] for n=1, 2, ..., M do

[0132] S T =EUN ( T )

[0133] loss=L ( S , GT )

[0134] end

[0135] End

[0136] It should be noted that through verification, it can be seen that the brain tumor image segmentation model provided by the present invention weighs the pros and cons between segmentation accuracy and computational overhead. By introducing the attention mechanism and efficient convolutional network into the model, the local spatial and multi-dimensional feature understanding capabilities are improved. A large number of experimental results on the BraTS 2021 dataset show that the computational cost of 3D-EffUNet is reduced by 80% while maintaining good accuracy. In addition, automatic mixed precision training is used to keep the computational cost low.

[0137] like Figure 5 As shown, a brain tumor image segmentation model construction system includes:

[0138] A model building module, used to build an initial brain tumor image segmentation model including an encoder and a decoder;

[0139] A sample image module is used to divide the preprocessed brain tumor images into a training set and a validation set;

[0140] A model training module, used to complete model training according to a training set, an initial brain tumor image segmentation model, and a preset loss;

[0141] The verification module is used to determine a brain tumor image segmentation model that meets preset conditions based on a verification set.

[0142] In actual application, the brain tumor image segmentation model construction system includes a model construction module, a sample image module, a model training module and a verification module. The functions of each module correspond to the corresponding steps and should have the same beneficial effects, which will not be repeated here.

[0143] In the embodiments provided in the present application, it should be understood that the disclosed methods and systems can be implemented in other ways. The system embodiments described above are merely schematic. For example, the division of modules is only a logical function division. There may be other division methods in actual implementation, such as: multiple modules or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0144] In addition, all functional modules in the embodiments of the present invention may be integrated into one processor, or each module may be a separate device, or two or more modules may be integrated into one device; each functional module in the embodiments of the present invention may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0145] A person of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by program instructions and related hardware. The aforementioned program instructions can be stored in a computer-readable storage medium. When the program instructions are executed, the steps of the above method embodiment are executed; and the aforementioned storage medium includes: a mobile storage device, a read-only memory (ROM), a magnetic disk or an optical disk, and other media that can store program codes.

[0146] It should be understood that the use of "system", "device", "unit" and / or "module" in this application is only a method for distinguishing different components, elements, parts, parts or assemblies at different levels. However, if other words can achieve the same purpose, the word can be replaced by other expressions.

[0147] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "a kind" and / or "the" do not refer to the singular, but also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements. The elements defined by the sentence "includes a..." do not exclude the existence of other identical elements in the process, method, commodity or device that includes the elements.

[0148] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features.

[0149] If a flow chart is used in the present application, the flow chart is used to illustrate the operations performed by the system according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, each step can be processed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or a certain step or several steps of operations can be removed from these processes.

[0150] The above is a detailed introduction to a method and system for constructing a brain tumor image segmentation model provided by the present invention. The above description of the disclosed embodiments enables professionals and technicians in the field to implement or use the present invention. Various modifications to these embodiments will be obvious to professionals and technicians in the field, and the general principles defined in this article can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown in this article, but will conform to the widest scope consistent with the principles and novel features disclosed in this article.

Claims

1. A method for constructing a brain tumor image segmentation model, characterized in that: The steps include: Build an initial brain tumor image segmentation model including an encoder and a decoder; The preprocessed brain tumor images were divided into training set and validation set; Complete model training according to the training set, the initial brain tumor image segmentation model and the preset loss; Determining a brain tumor image segmentation model that meets preset conditions according to the verification set; Wherein, the initial brain tumor image segmentation model further includes: a preprocessing convolution module; The preprocessing convolution module is used to stretch the channel length of brain tumor images of different modes; The encoder is used to compress the size of the brain tumor image to extract features between space and channels, and output the result; The decoder is used to segment the features of the brain tumor image according to the output result of the encoder; The encoder comprises: a plurality of efficient convolution modules; The efficient convolution module includes: a first point-by-point convolution layer, a channel-by-channel convolution layer, and a second point-by-point convolution layer; The channel-by-channel convolution layer is arranged between the first point-by-point convolution layer and the second point-by-point convolution layer; Each convolutional layer applies an activation function for group normalization; The encoder also includes: an attention module; The attention module is arranged between the channel-by-channel convolution layer and the second point-by-point convolution layer; The attention module is used to apply channel-space attention to the maximum representation in the channel-space field; The decoder comprises: a plurality of up-convolution modules and a Conv3D module; The upper convolution module is connected to the efficient convolution module and the Conv3D module respectively; The up-convolution module cooperates with the efficient convolution module to reconstruct the position features in the brain tumor image; The Conv3D module is used to obtain the confidence of different position features of the brain tumor image region according to the position features in the brain tumor image to segment the brain tumor image; The upper convolution module includes: a transposed three-dimensional convolution layer, a first three-dimensional convolution layer and a second three-dimensional convolution layer; The transposed three-dimensional convolution layer is used to concatenate the output result with the output result of the efficient convolution module as input data; The first three-dimensional convolution layer and the second three-dimensional convolution layer are used to reconstruct the position features in the brain tumor image according to the input data output; Each convolutional layer applies an activation function for group normalization; The model training is completed according to the training set, the initial brain tumor image segmentation model and the preset loss, specifically: Performing model training on the initial brain tumor image segmentation model according to the cross entropy loss, the DICE loss function, the preset trade-off factor and the training set to minimize the total loss; The model training formula is as follows: ； in, is the total loss, is the cross entropy loss between the brain tumor labels segmented by the model and the true brain tumor labels, is the DICE loss between the two, is a trade-off factor.

2. A brain tumor image segmentation model construction system, characterized in that: include: A model building module, used to build an initial brain tumor image segmentation model including an encoder and a decoder; A sample image module is used to divide the preprocessed brain tumor images into a training set and a validation set; A model training module, used to complete model training according to the training set, the initial brain tumor image segmentation model and a preset loss; A verification module, used to determine a brain tumor image segmentation model that meets preset conditions according to the verification set; Wherein, the initial brain tumor image segmentation model further includes: a preprocessing convolution module; The preprocessing convolution module is used to stretch the channel length of brain tumor images of different modes; The encoder is used to compress the size of the brain tumor image to extract features between space and channels, and output the result; The decoder is used to segment the features of the brain tumor image according to the output result of the encoder; The encoder comprises: a plurality of efficient convolution modules; The efficient convolution module includes: a first point-by-point convolution layer, a channel-by-channel convolution layer, and a second point-by-point convolution layer; The channel-by-channel convolution layer is arranged between the first point-by-point convolution layer and the second point-by-point convolution layer; Each convolutional layer applies an activation function for group normalization; The encoder also includes: an attention module; The attention module is arranged between the channel-by-channel convolution layer and the second point-by-point convolution layer; The attention module is used to apply channel-space attention to the maximum representation in the channel-space field; The decoder comprises: a plurality of up-convolution modules and a Conv3D module; The upper convolution module is connected to the efficient convolution module and the Conv3D module respectively; The up-convolution module cooperates with the efficient convolution module to reconstruct the position features in the brain tumor image; The Conv3D module is used to obtain the confidence of different position features of the brain tumor image region according to the position features in the brain tumor image to segment the brain tumor image; The upper convolution module includes: a transposed three-dimensional convolution layer, a first three-dimensional convolution layer and a second three-dimensional convolution layer; The transposed three-dimensional convolution layer is used to concatenate the output result with the output result of the efficient convolution module as input data; The first three-dimensional convolution layer and the second three-dimensional convolution layer are used to reconstruct the position features in the brain tumor image according to the input data output; Each convolutional layer applies an activation function for group normalization; The model training module is specifically used to perform model training on the initial brain tumor image segmentation model according to the cross entropy loss, the DICE loss function, the preset trade-off factor and the training set to minimize the total loss; The model training formula is as follows: ； in, is the total loss, is the cross entropy loss between the brain tumor labels segmented by the model and the true brain tumor labels, is the DICE loss between the two, is a weighing factor; Among them, is the total loss, is the cross entropy loss between the brain tumor label segmented by the model and the true brain tumor label, is the DICE loss between the two, and is the trade-off factor.

Images