Medical image segmentation method fused with multi-layer perceptron, terminal equipment and medium

Abstract

The invention relates to a medical image segmentation method fused with a multi-layer perceptron, terminal equipment and a medium, and the method comprises the steps: constructing a medical image segmentation model, employing a U-shaped encoder-decoder architecture, combining the advantages of convolution and MLP, and fusing the convolution and the MLP into each layer; a multi-axis and multi-window MLP module is designed, feature map information is captured from multiple windows and multiple scales, pictures do not need to be cut during input, the pictures of any size can be received, and a global receptive field always exists; a Conv-MCG module is developed, two outputs of a convolution module are subjected to feature cross fusion, and multi-path and multi-information interaction is achieved. According to the method, a relatively good segmentation result can be obtained on relatively small calculation time and space cost.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a medical image segmentation method, terminal equipment and medium integrating a multi-layer perceptron. Background Art

[0002] In recent years, convolutional neural networks (CNN) have made significant progress in computer vision applications, such as image classification, object detection, semantic segmentation, object tracking, image enhancement, etc. Thanks to the development of CNN, computer vision technology has been widely used in the field of medical image processing. Image semantic segmentation is an important part of medical image processing and can play an important role in computer-aided diagnosis and image-guided clinical surgery.

[0003] The most typical existing deep learning-based medical image segmentation network is U-Net, which consists of a symmetrical encoder-decoder and skip connections. In the encoder, a series of convolutional layers and continuous downsampling layers are used to increase the receptive field and extract more semantic features. Then, the decoder upsamples the extracted semantic features to obtain pixel-level semantic predictions, and fuses the shallow features from the encoder with the deep features from the decoder through skip connections to reduce the loss of semantic information caused by deep convolution and downsampling. With its excellent structural design, U-Net has achieved great success in medical image processing. Based on this network, many variant networks have been developed for various types of medical image segmentation. The excellent performance of these U-Net-based methods in medical image segmentation has proved that CNN has a strong ability to learn features. However, the inherent locality and weight sharing of the receptive field in the convolution operation make it difficult for U-Net-based methods to learn explicit global information and interact with long-range semantic information, which to a certain extent cannot meet the strict requirements of segmentation accuracy in the field of medical image segmentation.

[0004] Applying Transformer in computer vision can alleviate long-distance dependencies to a certain extent. At the same time, Transformer has a strong ability to model global relationships and has achieved good results in medical image analysis tasks. Existing studies have proposed Vision Transformer (ViT) to perform image recognition tasks, taking two-dimensional image blocks with position embedding as input, and achieved very good performance. Research based on ViT has also promoted the study of multi-layer perceptron (MLP) structures, such as MLP-Mixer, gMLP, and aMLP. In particular, MLP-Mixer is a network based entirely on MLP, which gives performance comparable to Transformer with less computation. Although existing Transformer- and MLP-based methods have proven to be very promising in image analysis tasks including medical image segmentation, there are still several daunting challenges:

[0005] (1) It only accepts fixed image sizes and needs to rigidly divide the image into fixed sizes, which may not be able to capture the fine-grained spatial detail information of the image.

[0006] (2) When applied to larger images, boundary artifacts will inevitably occur.

[0007] (3) Since the input is regarded as a one-dimensional sequence and only the modeling of global information is focused on at all stages, detailed localization information is lacking. When performing semantic analysis, the ability to locate the location of interest may be lacking.

[0008] These problems are the shortcomings of Transformer and MLP compared to CNN networks in extracting underlying features and visual structures. Summary of the invention

[0009] In order to solve the above problems, the present invention proposes a medical image segmentation method, terminal equipment and medium integrating a multi-layer perceptron.

[0010] The specific plan is as follows:

[0011] A medical image segmentation method integrating a multi-layer perceptron comprises the following steps:

[0012] S1: Collect medical image data to build a training set;

[0013] S2: Build a medical image segmentation model and train the model using the training set;

[0014] The network backbone of the model adopts a U-shaped network, including N encoders, N decoders and an output module; the encoder, decoder and output module each include three convolution blocks, an MLP cross-gating module and a splicing module;

[0015] In the encoder, data is input from the first convolution block, the output of the first convolution block is input to the second convolution block after downsampling, the output of the second convolution block is input to the third convolution block after upsampling, the output of the first convolution block and the output of the third convolution block are input to the MLP cross-gating module together, the output of the MLP cross-gating module is connected to the decoder through jump connection, and on the other hand, it is spliced ​​with the output of the second convolution block in the splicing module after downsampling, and the output of the splicing module is the output of the encoder;

[0016] In the decoder, the output data of the previous layer is input from the first convolution block. The output of the first convolution block is upsampled and input to the second convolution block together with the data of the jump connection of the encoder of the previous layer. The output of the second convolution block is downsampled and input to the third convolution block together with the jump connection data of the encoder of the corresponding layer. The output of the first convolution block and the output of the third convolution block are input to the MLP cross-gating module together. The output of the MLP cross-gating module is upsampled and spliced ​​with the output of the second convolution block in the splicing module. The output of the splicing module is the output of the decoder.

[0017] In the output module, the output data of the decoder is input from the first convolution block, the output of the first convolution block is directly input into the second convolution block, the output of the first convolution block and the output of the second convolution block are input into the MLP cross-gating module together, the output of the MLP cross-gating module is up-sampled and then spliced ​​with the data of the first-layer encoder jump connection in the splicing module, the output of the splicing module is used as the input of the third convolution block, and the output of the third convolution block is the model output result;

[0018] S3: Segment the medical image to be segmented using the trained model.

[0019] Furthermore, the MLP cross-gating module includes inputs of two paths. For each path, its input data is sequentially subjected to layer normalization, a fully connected layer, and a GELU activation function, and then input into a multi-axis multi-window MLP structure. The output of the multi-axis multi-window MLP structure is multiplied by the input of the multi-axis multi-window MLP structure. The multiplication result is sequentially subjected to a fully connected layer and Dropout activation, and the Dropout activation result is added to the input data of the path to obtain the output result of the path; the output results of the two paths are added to obtain the output result of the MLP cross-gating module.

[0020] Furthermore, the multi-axis multi-window MLP structure divides the input feature map into four heads according to the channels, and uses MLP for information fusion in each head, where two heads are sent to the local feature branch and the other two heads are sent to the global feature branch; in the local feature branch, the two heads are respectively sent to two local feature processing axes with different window sizes; in the global feature branch, the two heads are respectively sent to two global feature processing axes with different window sizes.

[0021] Furthermore, the convolution block consists of a combination of two 3×3 convolutional layers, a BN layer and a ReLU activation function layer.

[0022] Furthermore, step S1 also includes augmenting the collected medical image data by random rotation or hue, brightness, and cropping adjustments, and constructing a training set based on the augmented data.

[0023] A medical image segmentation terminal device integrating a multi-layer perceptron comprises a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method described above in the embodiment of the present invention are implemented.

[0024] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described above in an embodiment of the present invention are implemented.

[0025] The present invention adopts the above technical solution, which can achieve the best segmentation performance while keeping the learnable parameters of the model and the calculation cost small. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Shown is a flow chart of a method according to a first embodiment of the present invention.

[0027] Figure 2 Shown is a schematic diagram of the network structure of the model in this embodiment.

[0028] Figure 3 Shown is a schematic diagram of the structure of each module inside the model in this embodiment.

[0029] Figure 4 Shown is a schematic structural diagram of the MCG module in this embodiment.

[0030] Figure 5 Shown is a schematic diagram of the multi-axis multi-window MLP structure in this embodiment. DETAILED DESCRIPTION

[0031] To further illustrate various embodiments, the present invention provides drawings. These drawings are part of the disclosure of the present invention, which are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in this field should be able to understand other possible implementations and advantages of the present invention.

[0032] The present invention will now be further described with reference to the accompanying drawings and specific implementation methods.

[0033] Embodiment 1:

[0034] The embodiment of the present invention provides a medical image segmentation method integrating a multi-layer perceptron, such as Figure 1 As shown, the method comprises the following steps:

[0035] S1: Collect medical image data to build a training set.

[0036] The medical image data collected in this embodiment comes from the following three existing datasets:

[0037] Breast UltraSound Images (BUSI) is a medical image dataset of breast cancer scanned by ultrasound. The data collected by BUSI includes breast ultrasound images of women aged 25 to 75 years old, collected in 2018. The number of patients is 600 women. The dataset consists of 780 images, all of which are cropped to different sizes to remove unused and unimportant boundaries from the images. The pictures are in PNG format and are divided into three categories, namely normal, benign, and malignant. Each image has its own Ground Truth (GT). In this example, benign and malignant tumor images are selected, and a total of 647 images are uniformly adjusted to 256×256 RGB images.

[0038] The International Skin Imaging Collaboration (ISIC 2018) is the world's largest skin image analysis challenge dataset. The dataset contains 2594 images in three different categories, including 20.0% melanoma, 72.0% nevus, and 8.0% oleoresinus keratosis. The dataset consists of images of various resolutions. In this example, all images are resized to 512×512 RGB images.

[0039] CVC-ClinicDB is a colonoscopy image dataset consisting of endoscopic images. These images are extracted from colonoscopy video sequences. The ground truth image consists of a mask corresponding to the area covered by polyps in the image. Since the segmentation task only considers images of polyps, a total of 612 images from 31 colonoscopy sequences are obtained. In this example, all images are resized to 256×256 RGB images.

[0040] In order to increase the diversity of training samples and make the trained model have stronger generalization ability, the collected original medical image data is also augmented in this embodiment, and the image is randomly rotated and the hue, brightness, cropping and other random adjustments are made.

[0041] S2: Build a medical image segmentation model and train the model using the training set.

[0042] For medical image segmentation, many research results show that the U-Net network architecture is a very competitive infrastructure. Therefore, the main framework of the medical image segmentation model (abbreviated as MsMCG below for convenience of description) in this embodiment still adopts the U-Net architecture. Figure 2 As shown. The U-shaped architecture has great advantages in extracting underlying features, but its local receptive field and long-distance weak dependence still hinder the accuracy of the segmentation results. In addition to the advantages of U-Net, an excellent network should also add MLP to increase the global receptive field of the network. Based on this idea, a Conv-MLP encoder module and a Conv-MLP decoder module are designed in this embodiment to build the encoder and decoder of the MsMCG network. The MLP Cross Gating (MCG) in the Conv-MLP encoder module and the Conv-MLP decoder module is used to fuse the information of the two convolutional layers, and the fine local features extracted from the convolution information are obtained through MLP to obtain the global receptive field. The output of the MCG in the Conv-MLP encoder module is jump-connected to the decoder, and each MCG submodule is connected to different Conv-MLP decoder modules twice. The proposed architecture adopts an end-to-end training strategy, relying on the hybrid model design of each block to adaptively obtain local and global receptive fields at each layer. Conv is used for local and MLP is used for long-distance interaction to make full use of every information.

[0043] The core building blocks of MsMCG are Conv-MLP encoder and Conv-MLP decoder, which are the backbone of the U-type network. Figure 2As shown, MsMCG includes N encoders, N decoders and an output module, and in this embodiment, N is preferably set to 4. Convolutional block design is used in each module, which consists of two Conv3×3 (3×3 convolutional layers) + BN + ReLU.

[0044] In the Conv-MLP encoder (such as Figure 3 (a) shows that the U-Net network encoder design concept is still used, but considering the problem of information loss during the downsampling process, the convolution block is used to downsample and then upsample, so that the information of the next layer can be obtained while losing as little information as possible. That is, after the data is input from the first convolution block, the output of the first convolution block is input to the second convolution block after downsampling, and the output of the second convolution block is input to the third convolution block after upsampling. In addition, the outputs of two convolution blocks (the first convolution block and the third convolution block) in the same layer will enter the MLP Cross Gating (MCG) module, which can cross-fuse the information of the convolution blocks. The output of the MLP Cross Gating module will be used for jump connection with the decoder. At the same time, the feature map size of the output of the MLP Cross Gating module is doubled after downsampling, and it is spliced ​​with the output of the convolution block (the second convolution block) obtained by downsampling in the previous convolution operation in the channel dimension (splicing module, Concat) for the operation of the next module.

[0045] The decoder design of the U-Net network is also used in the Conv-MLP decoder to gradually restore the resolution of the image and merge it with the decoder information. In this embodiment, the convolution block is used to upsample and then downsample, which can obtain information with a higher resolution in the previous layer. This design makes the information in the current layer more forward-looking and can perceive more detailed information. That is, after the output data of the previous layer of the decoder (the output data of the previous layer decoder or the output data of the last layer encoder) is input from the first convolution block, the output of the first convolution block is upsampled and input together with the jump connection data of the previous layer encoder to the second convolution block, and the output of the second convolution block is downsampled and input together with the jump connection data of the input current layer encoder to the third convolution block. The output of the first convolution block and the output of the third convolution block are input together to the MLP cross-gating module. The output of the MLP cross-gating module is upsampled and spliced ​​with the output of the second convolution block in the splicing module. The output result of the splicing module is the output of the decoder. The structure of the Conv-MLP decoder is as follows: Figure 3 As shown in (b) in the figure, similar to the design of the encoder, the convolutional blocks in the decoder of each layer are simultaneously fed into the MCG module twice and cross-fused through the MCG module.

[0046] In order to better integrate the information of convolution and MLP, an MLP cross-gating (MCG) module is designed in this embodiment, such as Figure 4 As shown in Figure 2. This module can well preserve the ability of convolution operation to extract local and fine features, while also expanding the feature map to the global receptive field. The MCG module receives inputs from two paths, which are the outputs of two convolution blocks. Suppose the inputs of the two paths are U1 and V1, respectively, U1, V1∈R H×W×C , the input is operated on:

[0047] U2=σ(Ds(LN(U1)))V2=σ(Ds(LN(V1))) (1)

[0048] Among them, LN (LayerNorm) represents layer normalization, Ds (Dense) represents the fully connected layer, and σ represents the activation function GELU. After this series of operations, U2 and V2 are obtained, and then enter the multi-axis multi-window MLP structure respectively. First, the processed data is divided into:

[0049] (L1,L2,G1,G2)=S(σ(Ds(LN(X)))) (2)

[0050] Here, S means splitting the input feature map into four heads for operation. These four heads can be used for operations on different axes, as shown in formula (3).

[0051] M(X)=Do(Ds([Local1(L1),Local2(L2),Global1(G1),Global2(G2)])) (3)

[0052] Here, the square brackets [] represent the concatenation operation, and M(X) represents the output of the multi-axis multi-window MLP module, which are M u and M v , used for the fusion of two paths.

[0053] U3=U2⊙M v (4)

[0054] V3=V2⊙M u (5)

[0055] Where ⊙ represents the product operation of the corresponding elements, and the output of each path is obtained by residual connection with the input.

[0056] U4=U1+(ρ(Ds(U3))) (6)

[0057] V4=V1+(ρ(Ds(V3))) (7)

[0058] Where ρ represents Dropout activation. In order to obtain two inputs and one output that meet the design, the two paths are added in this embodiment, so:

[0059] M=U4+V4 (8)

[0060] In 2021, researchers in this field proposed the idea of ​​sparse self-attention operation, which performs different forms of sparse self-attention on the two axes of the block image to capture local and global information in parallel. However, the test image input to this module needs to be of a fixed size, which can easily cause boundary blur and boundary artifacts, and is not friendly to model training of large images. Based on this, a multi-axis multi-window MLP framework is proposed in this embodiment, which establishes a four-axis image information fusion framework, such as Figure 5 shown.

[0061] The multi-axis multi-window MLP structure divides the input feature map into four heads according to the channel, and uses MLP to fuse information in each head, two of which are sent to the local feature branch and the other two heads are sent to the global feature branch.

[0062] In the local feature branch, the two heads are fed into two feature processing axes respectively, and the input tensor size of each axis is (H, W, C / 4). At the same time, one of the axes is divided into H / b×W / b non-overlapping windows according to the window size [b×b]. In this way, the dimension of the input tensor changes to obtain a tensor of (H / b×W / b, b×b, C / 4). The other axis is divided into H / 2b×W / 2b non-overlapping windows according to the window size [2b×2b], and the dimension of the input tensor changes to obtain a tensor of (H / 2b×W / 2b, b×b, C / 4).

[0063] In the global feature branch, one of the axes is divided into b×b non-overlapping windows of size H / b×W / b according to the grid size [b×b], and each window size is H / b×W / b. After reshaping the dimension, a tensor of (b×b, H / b×W / b, C / 4) is obtained. Similarly, the other axis is divided into 2b×2b windows of size H / 2b×W / 2b according to the grid size [2b×2b], and a tensor of (2b×2b, H / 2b×W / 2b, C / 4) is obtained after reshaping the dimension.

[0064] exist Figure 5In , we take b = 2, the upper half represents the local branch, and the lower half represents the global branch. It can be seen from the figure that after the input feature map passes through the LayerNorm, Dense and GELU layers, it is divided into 4 axes by channel, half of which enter the local branch head and the other half enter the global branch head. In both the local branch and the global branch, windows of different sizes are used for division, and MLP is used for different degrees of information fusion. The local branch corresponds to the local feature fusion, and the global branch corresponds to the global feature fusion. The blocks of different colors in the figure represent the fusion of information in the window. Finally, the processed 4 axes are spliced ​​together and output through the Dense and Dropout layers. The design of this module is completely end-to-end from input to output, does not require a specific size, and does not need to crop the image when used, thus avoiding some bad effects caused by cropping. In addition, the complexity of the model is linear and the amount of calculation is not large.

[0065] In the output (Conv-MLP) module (such as Figure 3 In (c), after the output data of the decoder is input from the first convolution block, the output of the first convolution block is directly input into the second convolution block, and the output of the first convolution block and the output of the second convolution block are input into the MLP cross-gating module together. The output of the MLP cross-gating module is upsampled and then spliced ​​with the data of the first-layer encoder jump connection in the splicing module. The output of the splicing module is input into the third convolution block, and the output of the third convolution block is the model output result.

[0066] S3: Segment the medical image to be segmented using the trained model.

[0067] In order to integrate the advantages of convolution and MLP, the embodiment of the present invention proposes a new deep convolutional neural network architecture based on U-shaped structure for medical image segmentation. The core idea is to integrate the respective advantages of convolution and MLP, and integrate the two into each layer of the encoder and decoder. At the same time, feature map information can be captured from multiple windows and multiple dimensions, and this network always has a global receptive field. The proposed MsMCG architecture effectively builds the network with a small amount of parameters and computation. Experimental results show that the proposed architecture outperforms the most advanced baselines on various benchmarks for segmentation of biomedical images from different types. In addition, our network has achieved very excellent results for image segmentation under complex backgrounds.

[0068] The embodiments of the present invention have the following technical effects:

[0069] (1) proposed a general framework for medical image segmentation, called MsMCG. Using a U-shaped encoder-decoder architecture, it combines the advantages of convolution and MLP and integrates them into each layer.

[0070] (2) A multi-axis and multi-window MLP module is designed to capture local and global information from multiple dimensions and distribute it equally to the obtained branches by channel. The information is mixed according to different mechanisms on each axis. Its input does not require image cropping, can accept images of any size, and always has a global receptive field.

[0071] (3) A Conv-MCG module was developed to cross-fuse the features of the two outputs of the convolution module, which is also a multi-path and multi-information interaction.

[0072] (4) The designed network has achieved some results in light weight while improving segmentation accuracy. Based on the linear relationship between the computational complexity of the MLP module and the size of the input feature map, the feature maps output by the two convolutional blocks are input into the MCG module to obtain local and global receptive fields at each layer for deeper information fusion. Compared with most existing networks, the performance of indicators such as the number of model parameters and running speed is better. That is, better segmentation results can be achieved with less computing time and space costs.

[0073] Embodiment 2:

[0074] The present invention also provides a medical image segmentation terminal device integrating a multi-layer perceptron, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps in the above-mentioned method embodiment of the first embodiment of the present invention when executing the computer program.

[0075] Further, as an executable solution, the medical image segmentation terminal device fused with a multi-layer perceptron can be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The medical image segmentation terminal device fused with a multi-layer perceptron may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the composition structure of the medical image segmentation terminal device fused with a multi-layer perceptron is only an example of a medical image segmentation terminal device fused with a multi-layer perceptron, and does not constitute a limitation on the medical image segmentation terminal device fused with a multi-layer perceptron. It may include more or less components than the above, or a combination of certain components, or different components. For example, the medical image segmentation terminal device fused with a multi-layer perceptron may also include input and output devices, network access devices, buses, etc., and the embodiment of the present invention does not limit this.

[0076] Further, as an executable solution, the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the medical image segmentation terminal device fused with a multi-layer perceptron, and uses various interfaces and lines to connect various parts of the entire medical image segmentation terminal device fused with a multi-layer perceptron.

[0077] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the medical image segmentation terminal device fused with the multi-layer perceptron by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and an application required for at least one function; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0078] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method in the embodiment of the present invention are implemented.

[0079] If the module / unit integrated in the medical image segmentation terminal device integrating the multi-layer perceptron is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory) and software distribution medium, etc.

[0080] Although the present invention has been specifically shown and described in conjunction with the preferred embodiments, it should be understood by those skilled in the art that various changes may be made to the present invention in form and details without departing from the spirit and scope of the present invention as defined by the appended claims, all of which are within the scope of protection of the present invention.

Claims

1. A medical image segmentation method integrating a multi-layer perceptron, characterized in that: The following steps are involved: S1: Collect medical image data to build a training set; S2: Build a medical image segmentation model and train the model using the training set; The network backbone of the model adopts a U-shaped network, including N encoders, N decoders and an output module; the encoder, decoder and output module each include three convolution blocks, an MLP cross-gating module and a splicing module; In the encoder, data is input from the first convolution block, the output of the first convolution block is input to the second convolution block after downsampling, the output of the second convolution block is input to the third convolution block after upsampling, the output of the first convolution block and the output of the third convolution block are input to the MLP cross-gating module together, the output of the MLP cross-gating module is connected to the decoder through jump connection, and on the other hand, it is spliced ​​with the output of the second convolution block in the splicing module after downsampling, and the output of the splicing module is the output of the encoder; In the decoder, the output data of the previous layer is input from the first convolution block. The output of the first convolution block is upsampled and input to the second convolution block together with the data of the jump connection of the encoder of the previous layer. The output of the second convolution block is downsampled and input to the third convolution block together with the jump connection data of the encoder of the corresponding layer. The output of the first convolution block and the output of the third convolution block are input to the MLP cross-gating module together. The output of the MLP cross-gating module is upsampled and spliced ​​with the output of the second convolution block in the splicing module. The output of the splicing module is the output of the decoder. In the output module, the output data of the decoder is input from the first convolution block, the output of the first convolution block is directly input into the second convolution block, the output of the first convolution block and the output of the second convolution block are input into the MLP cross-gating module together, the output of the MLP cross-gating module is up-sampled and then spliced ​​with the data of the first-layer encoder jump connection in the splicing module, the output of the splicing module is used as the input of the third convolution block, and the output of the third convolution block is the model output result; S3: Segment the medical image to be segmented using the trained model.

2. The medical image segmentation method integrating multi-layer perceptron according to claim 1 is characterized in that: The MLP cross-gating module includes inputs of two paths. For each path, its input data is sequentially normalized, fully connected, and activated by the GELU function before being input into the multi-axis multi-window MLP structure. The output of the multi-axis multi-window MLP structure is multiplied by the input of the multi-axis multi-window MLP structure. The multiplication result is sequentially passed through the fully connected layer and Dropout activation. The Dropout activation result is added to the input data of the path to obtain the output result of the path. The output results of the two paths are added to obtain the output result of the MLP cross-gating module.

3. The medical image segmentation method integrating multi-layer perceptron according to claim 1, characterized in that: The multi-axis multi-window MLP structure divides the input feature map into four heads according to the channel, and uses MLP to fuse information in each head, two of which are sent to the local feature branch, and the other two heads are sent to the global feature branch; in the local feature branch, the two heads are respectively sent to two local feature processing axes with different window sizes; in the global feature branch, the two heads are respectively sent to two global feature processing axes with different window sizes.

4. The medical image segmentation method integrating multi-layer perceptron according to claim 1 is characterized in that: The convolution block consists of a combination of two 3×3 convolutional layers, a BN layer and a ReLU activation function layer.

5. The medical image segmentation method integrating multi-layer perceptron according to claim 1, characterized in that: Step S1 also includes augmenting the collected medical image data by random rotation or hue, brightness, and cropping adjustments, and constructing a training set based on the augmented data.

6. A medical image segmentation terminal device integrating a multi-layer perceptron, characterized in that: The method comprises a processor, a memory and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of any one of the methods of claims 1 to 5 when executing the computer program.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images