A breast segmentation model training method and device for breast ultrasound image lesion area recognition

Abstract

The present application discloses a breast segmentation model training method and device for identifying lesion areas in breast ultrasound images. The breast segmentation model training method for identifying lesion areas in breast ultrasound images includes: obtaining a breast patient ultrasound image dataset, wherein the breast patient ultrasound image dataset includes at least one breast patient ultrasound image; obtaining a breast segmentation model, wherein the breast segmentation model includes an MGSA module, and the MGSA module is used to obtain a weighted guided feature map based on the original image; and training the breast segmentation model using the breast patient ultrasound image dataset to obtain a trained breast segmentation model. The method of the present application can provide more reliable support for auxiliary diagnosis of breast ultrasound images. For problems such as fuzzy boundaries in breast ultrasound image lesion segmentation and insufficient capture of contextual information, it can ensure higher segmentation accuracy and stronger robustness.

Description

Technical Field

[0001] The present application relates to the field of image recognition technology, and in particular to a breast segmentation model training method for identifying lesion areas in breast ultrasound images and a breast segmentation model training device for identifying lesion areas in breast ultrasound images. Background Art

[0002] Breast cancer has become a major threat to women's health worldwide due to its high morbidity and mortality rates. Despite the continuous advancement of modern medical diagnostic and treatment technologies, the early diagnosis and treatment of breast cancer still face huge challenges. Breast ultrasound imaging, as a non-invasive, convenient and economical diagnostic tool, has become an important means of screening and early diagnosis of breast cancer. With the development of medical technology, especially the continuous innovation of medical imaging technology, the quality of breast ultrasound imaging has been significantly improved and has been widely used in clinical practice. However, the low contrast, noise interference, and blurred boundaries between lesions and background in breast ultrasound images pose significant challenges to automated image segmentation. These challenges limit the performance of computer-aided diagnosis (CAD) systems, thereby affecting the accuracy and efficiency of tumor detection.

[0003] Traditionally, tumor segmentation in breast ultrasound images relies on doctors manually labeling lesion regions. While accurate, this method not only requires extensive experience and expertise, but is also extremely inefficient when dealing with large numbers of images, failing to meet growing clinical needs. To improve diagnostic efficiency and reduce the workload of doctors, automated tumor segmentation has become a hot topic of research. The rapid development of deep learning technology in recent years, particularly the successful application of convolutional neural networks (CNNs) in medical image analysis, has provided a new solution for automated segmentation of breast ultrasound images.

[0004] Therefore, it is hoped that there will be a technical solution to solve or at least alleviate the above-mentioned deficiencies in the prior art. Summary of the Invention

[0005] The object of the present invention is to provide a breast segmentation model training method for identifying lesion areas in breast ultrasound images to at least solve one of the above technical problems.

[0006] The present invention provides the following solutions:

[0007] According to one aspect of the present invention, a breast segmentation model training method for identifying lesion areas in breast ultrasound images is provided. The breast segmentation model training method for identifying lesion areas in breast ultrasound images comprises:

[0008] Acquire a breast patient ultrasound image dataset, where the breast patient ultrasound image dataset includes at least one breast patient ultrasound image;

[0009] Acquire a breast segmentation model, wherein the breast segmentation model includes a MGSA module, and the MGSA module is used to obtain a weighted guided feature map according to the original image;

[0010] The breast segmentation model is trained using the breast patient ultrasound image dataset to obtain a trained breast segmentation model.

[0011] Optionally, the breast segmentation model further includes a channel shuffling attention module, which is used to perform dimension permutation, inter-channel dependency capture, channel shuffling operation, and spatial attention enhancement on the weighted guided feature map, so as to obtain the final output feature map.

[0012] Optionally, the breast segmentation model further includes a DySample upsampling module, which is used for pixel-level spatial rearrangement of the final output feature map to use dynamically generated offsets to determine the positions of upsampling points, which reduces the neglect of positional relationships and the disordered shifting of offsets.

[0013] Optionally, the MGSA module includes:

[0014] A feature weight conversion module is used to perform Otsu binarization on the original image, segment the foreground and background, and concatenate the edge feature map obtained by the Scharr operator and the reverse feature map obtained by reverse attention with the decoded feature map output by the decoder. The concatenated feature map is then converted into feature weights using Sigmoid.

[0015] A product module is used to multiply the encoding feature map obtained by the encoding feature with the feature weight, so as to obtain a weighted guide feature map.

[0016] Optionally, the feature weight conversion module obtains the feature weight by the following formula:

[0017] Guide 1i =σ(Scharr(Pred i ))

[0018] Guide 2i =σ(Otsu(O ri ))

[0019] Guide 3i =1-σ(Pred i );

[0020] Among them, Guide 1i Guide is the edge guide feature weight, 2iGuide is the background guide feature weight, 3i To guide the feature weights in reverse attention, Pred i To decode the predicted feature map, O ri is the original image.

[0021] Optionally, the product module obtains a weighted guided feature map using the following formula:

[0022] Guide i =Cat(Guide 1i ×F ei )(Guide 2i ×F ei )(Guide 3i ×F ei );in,

[0023] Among them, Guide i Guide is a weighted guide feature map 1i Guide is the edge guide feature weight, 2i Guide is the background guide feature weight, 3i To guide the feature weights in reverse attention, Pred i To decode the predicted feature map, F ei is the encoding feature map.

[0024] Optionally, the step of performing dimension permutation on the weighted guided feature map and capturing inter-channel dependencies includes:

[0025] The weighted guided feature map contains multiple channels, and the spatial size of each channel is H×W. The weighted guided feature map first undergoes dimension permutation from C×H×W to W×H×C, and then captures the inter-channel dependency through a two-layer multilayer perceptron (MLP). The first layer of MLP reduces the number of channels to \(1 / 4\) times the original, and then introduces nonlinearity through the ReLU activation function. The second layer of MLP restores the number of channels to the original dimension. Finally, the inverse permutation is performed to restore it to C×H×W, and the channel attention map is generated through the Sigmoid activation function. The input feature map and the channel attention map are element-wise multiplied to obtain an enhanced feature map.

[0026] Optionally, the channel shuffling operation is specifically as follows:

[0027] The enhanced feature maps are divided into (4) groups, each group contains (C / 4) channels, and the grouped feature maps are transposed to disrupt the channel order within each group. Subsequently, the disrupted feature maps are restored to their original shape (C×H×W) to obtain the shuffled feature maps.

[0028] Optionally, the performing spatial attention enhancement includes:

[0029] The shuffled feature map passes through a 7x7 convolution layer to reduce the number of channels to (1 / 4) times the original size. Then, it undergoes batch normalization and ReLU activation function for nonlinear transformation. Then, it passes through a second 7x7 convolution layer to restore the number of channels to the original dimension C. It then passes through a batch normalization layer and finally, a sigmoid activation function to generate a spatial attention map.

[0030] The weighted guided feature map and the channel attention map are multiplied element-wise to obtain the final output feature map.

[0031] The present application also provides a breast segmentation model training device for identifying lesion areas in breast ultrasound images, characterized in that the breast segmentation model training device for identifying lesion areas in breast ultrasound images comprises:

[0032] A training set acquisition module, wherein the training set acquisition module is used to acquire a breast patient ultrasound image dataset, wherein the breast patient ultrasound image dataset includes at least one breast patient ultrasound image;

[0033] A breast segmentation model acquisition module, wherein the breast segmentation model acquisition module is used to obtain a breast segmentation model, wherein the breast segmentation model includes a MGSA module, and the MGSA module is used to obtain a weighted guided feature map based on the original image;

[0034] A training module is used to train the breast segmentation model using the breast patient ultrasound image dataset, thereby obtaining a trained breast segmentation model.

[0035] This application addresses challenges such as noise interference and blurred boundaries encountered during the segmentation of breast ultrasound images (BUS). A breast segmentation model training method for identifying lesion regions in BUS images is proposed. This method can be widely applied in complex clinical environments, providing more reliable support for auxiliary diagnosis of BUS images. It can ensure higher segmentation accuracy and greater robustness for addressing issues such as blurred boundaries and insufficient contextual information captured during BUS lesion segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 1 is a flow chart of a breast segmentation model training method for identifying lesion areas in breast ultrasound images in one embodiment of the present application;

[0037] Figure 2 This is a diagram of the breast segmentation model architecture in one embodiment of the present application.

[0038] Figure 3 This is the MGSA module architecture diagram.

[0039] Figure 4 The visualization results are compared with the existing methods.

[0040] Figure 5 Visualization results of the ablation experiment. DETAILED DESCRIPTION

[0041] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all 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.

[0042] like Figure 1 The breast segmentation model training method for identifying lesion areas in breast ultrasound images shown includes:

[0043] Step 1: Acquire a breast patient ultrasound image dataset, where the breast patient ultrasound image dataset includes at least one breast patient ultrasound image;

[0044] Step 2: Obtain a breast segmentation model, wherein the breast segmentation model includes a MGSA module, and the MGSA module is used to obtain a weighted guided feature map according to the original image;

[0045] Step 3: Training the breast segmentation model using the breast patient ultrasound image dataset to obtain a trained breast segmentation model.

[0046] This application addresses challenges such as noise interference and blurred boundaries encountered during the segmentation of breast ultrasound images (BUS). A breast segmentation model training method for identifying lesion regions in BUS images is proposed. This method can be widely applied in complex clinical environments, providing more reliable support for auxiliary diagnosis of BUS images. It can ensure higher segmentation accuracy and greater robustness for addressing issues such as blurred boundaries and insufficient contextual information captured during BUS lesion segmentation.

[0047] In this embodiment, step 1: obtaining a breast patient ultrasound image dataset, wherein the breast patient ultrasound image dataset includes at least one breast patient ultrasound image, can be performed by the following method:

[0048] Prepare a dataset of breast ultrasound images. This dataset consists of ultrasound images of breast cancer patients at Peking Union Medical College Hospital from January 2021 to July 2023, categorized as HER2-low-expressing and HER2-over-expressing. First, label the lesion regions in both datasets. After labeling, convert all the labeled JSON files into Cocoa-formatted txt files using labelme2coco for subsequent training.

[0049] 2) The breast ultrasound image dataset was divided into a training set, a test set, and a validation set in a ratio of 8:1:1. The training set was used to train the neural network, the test set was used to observe the training effect and determine the training termination time, and the validation set was used to verify the changes in training indicators after a certain period of time.

[0050] See also Figure 2 as well as Figure 3 In this embodiment, the breast segmentation model adopts U 2 -Net is used as the baseline for our model building, U 2 The -Net network architecture is a U-Net nested U-Net architecture. It consists of six encoders and five decoders. Next, the MGSA module is integrated between the encoders and decoders to guide background and edge feature learning for improved breast BUS image segmentation. The MGSA module utilizes the Otsu algorithm and the Scharr operator.

[0051] The MGSA module performs Otsu binarization on the original image to segment the foreground and background. It then performs Sigmoid processing on the edge feature map obtained by the Scharr operator and the reverse feature map obtained by reverse attention with the decoded prediction feature map output by the decoder, and converts them into corresponding guided feature weights:

[0052] Guide 1i =σ(Scharr(Pred i ));

[0053] Guide 2i =σ(Otsu(O ri ));

[0054] Guide 3i =1-σ(Pred i );

[0055] Among them, Guide 1i Guide is the edge guide feature weight, 2i Guide is the background guide feature weight, 3i To guide the feature weights in reverse attention, Pred i To decode the predicted feature map, Ori is the original image.

[0056] Multiply the above guide feature weights by the encoded feature map weights obtained by the encoder to obtain the weighted guide feature map Guide i :

[0057] Guide i =Cat(Guide 1i ×F ei )(Guid 2i ×F ei )(Guid 3i ×F ei )

[0058] Among them, F ei is the encoding feature map.

[0059] Channel shuffling attention strategy:

[0060] This part designs a global channel spatial attention mechanism, which integrates channel attention, channel permutation, and spatial attention to capture global dependencies. The guide feature map is first fed into the channel attention submodule and then into the spatial attention submodule. In the channel attention submodule, the guide feature map is first dimensionally permuted from C×H×W to W×H×C. Then, a two-layer multi-layer perceptron (MLP) is used to capture the inter-channel dependencies. The first layer of MLP reduces the number of channels to (1 / 4) times the original, then introduces nonlinearity through the ReLU activation function, and then the second layer of MLP restores the number of channels to the original dimension. Finally, the inverse permutation is performed to restore it to C×H×W, and the channel attention map is generated through the Sigmoid activation function. The input feature map and the channel attention map are element-wise multiplied to obtain the enhanced feature map:

[0061]

[0062] Among them F channel is the enhanced feature map, σ is the Sigmoid function, Represents element-wise multiplication.

[0063] To further mix and share information, a channel shuffle operation is applied. The enhanced feature maps are divided into (4) groups, each containing (C / 4) channels. The grouped feature maps are transposed to disrupt the order of channels within each group. Subsequently, the shuffled feature maps are restored to their original shape (C×H×W). This method can better mix feature information and enhance feature expression capabilities.

[0064] F shuffle =ChanelShuffle(F channel );

[0065] Among them F shuffle is the shuffled feature map.

[0066] In the spatial attention submodule, the input feature map passes through a 7x7 convolutional layer, reducing the number of channels to (1 / 4) the original number. It then undergoes batch normalization and a ReLU activation function for nonlinear transformation. Next, a second 7x7 convolutional layer restores the number of channels to the original dimension C, followed by another batch normalization layer. Finally, a sigmoid activation function is used to generate the spatial attention map. The shuffled feature map and the spatial attention map are element-wise multiplied to produce the final output feature map.

[0067]

[0068] Among them, F spatial It is the feature map after spatial attention. The final output feature map contains the enhanced features after channel attention, channel shuffling and spatial attention.

[0069] Dysample upsampling:

[0070] DySample upsampling is introduced into the network architecture to replace the original upsampling method. This improvement aims to improve the accuracy and robustness of breast ultrasound image segmentation by leveraging dynamic sampling and local awareness. DySample upsampling optimizes the upsampling process by adaptively adjusting the sampling position and offset range. In DySample, pixel-level spatial rearrangement of the feature map is performed to determine the location of the upsampling points using dynamically generated offsets, which reduces the neglect of positional relationships and the disordered shifting of offsets. This design enables more accurate feature recovery and avoids the artifacts and information loss common in traditional upsampling methods. Based on the form of the distance factor and the offset generation method, this method adopts a PL-style dynamic distance factor. This method first disrupts the pixel alignment of the input feature map through a pixel shuffling operation before applying the offset. It then generates offsets in each local region, which are subsequently used for upsampling.

[0071] Residual Connection:

[0072] We modified the normal convolutional connections to incorporate contextual information across channels, improving the model's segmentation performance. Specifically, we added residual connections to the network's inner layers, including convolution layers 3-8 of encoder 1, 3-7 of encoder 2, 3-6 of encoder 3, and 3-5 of encoder 5. The outputs from the convolutions, batch normalization layers, and ReLU activations are superimposed on the original input, preserving more original image information while minimizing interference during decoding.

[0073] In this embodiment, the present application further includes the following steps:

[0074] Fine-tune the breast segmentation model. The specific steps are as follows:

[0075] Our method is implemented in the PyTorch framework, using Python version 3.11 and PyTorch version 2.3.1. The network is optimized using the Adam optimizer with an initial learning rate of 10−4 and weight decay set to 10−7. All input images are resized to 288×288, and all experiments are conducted on an NVIDIA GeForce GTX 4060. Each model is trained for 100 epochs with a batch size of 4.

[0076] Verify the performance of the breast segmentation model and test the detection results. The specific steps are as follows:

[0077] 1) Compare our method with other existing segmentation methods. The experimental results are shown in Table 1.

[0078] Table 1 Comparison with advanced methods

[0079]

[0080] 2) To further demonstrate the superiority of our model, we conducted ablation experiments. We compared our proposed neural network model with the baseline, the baseline with residual connections, the baseline with Dysample upsampling, the baseline with the MGSA module, and the baseline with the MGSA module and Dysample. The ablation results for our proposed neural network model are shown in Table 2.

[0081] Table 2 Ablation experiment

[0082]

[0083] In comparative experiments with advanced methods, our model achieved improvements of 2.30, 1.28, 1.36, and 0.83 in the Jaccard, Precision, Sensitivity, and Dice metrics, respectively, on the HER2 underexpression dataset. On the HER2 overexpression dataset, our model achieved improvements of 1.44, 0.90, 2.89, and 1.99 in the Jaccard, Precision, Sensitivity, and Dice metrics, respectively. Ablation experiments also revealed the contributions of different modules or components in the final model.

[0084] 3) The visual results obtained by the test are as follows Figure 4 、 Figure 5 As shown. Among them, Figure 4 To visualize the results compared with the advanced methods, Figure 5 Visualize the results of the ablation experiment.

[0085] The present application also provides a breast segmentation model training device for identifying lesion areas in breast ultrasound images. The breast segmentation model training device for identifying lesion areas in breast ultrasound images includes a training set acquisition module, a breast segmentation model acquisition module, and a training module, wherein:

[0086] The training set acquisition module is used to acquire a breast patient ultrasound image dataset, where the breast patient ultrasound image dataset includes at least one breast patient ultrasound image;

[0087] The breast segmentation model acquisition module is used to acquire a breast segmentation model, wherein the breast segmentation model includes a MGSA module, and the MGSA module is used to acquire a weighted guided feature map according to the original image;

[0088] The training module is used to train the breast segmentation model using the breast patient ultrasound image dataset, thereby obtaining a trained breast segmentation model.

[0089] In response to the challenges of noise interference and blurred boundaries encountered in the segmentation process of breast ultrasound images (BUS), the present invention proposes a method for segmenting the lesion area in breast ultrasound images based on breast segmentation technology based on deep learning neural networks. This method can be widely used in complex clinical environments and provide more reliable support for auxiliary diagnosis of breast ultrasound images. For problems such as blurred boundaries and insufficient capture of contextual information in breast ultrasound image lesion segmentation, it can ensure higher segmentation accuracy and stronger robustness. The present invention is achieved through the following technical solutions, which are specifically:

[0090] The present invention uses the algorithm of deep convolutional neural network to design a network model based on U2-Net. The model takes breast ultrasound image (BUS) as the research object, extracts image features by using deep convolutional neural network, and then enhances the key information in the image through multi-guided attention module, combining the advantages of traditional image processing technology and deep learning to achieve high-precision segmentation of lesion area. The model includes an encoder, a decoder and a multi-guided channel shuffle attention (MGSA) module, which improves the processing ability of image edges and details. At the same time, the original bilinear upsampling is replaced by DySample upsampling, so that the model can obtain clearer areas and edge details. The method shows excellent performance in breast ultrasound image segmentation, has high accuracy and strong robustness, and is suitable for practical applications in clinical diagnosis.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A breast segmentation model training method for identifying lesion areas in breast ultrasound images, characterized in that: The breast segmentation model training method for breast ultrasound image lesion area recognition includes: Acquire a breast patient ultrasound image dataset, where the breast patient ultrasound image dataset includes at least one breast patient ultrasound image; Obtaining a breast segmentation model, wherein the breast segmentation model includes a MGSA module, an encoder, and a decoder, wherein the MGSA module is used to obtain a weighted guided feature map according to the original image; Training the breast segmentation model using the breast patient ultrasound image dataset to obtain a trained breast segmentation model; The breast segmentation model further includes a channel shuffling attention module, which is used to perform dimension permutation, inter-channel dependency capture, channel shuffling operation, and spatial attention enhancement on the weighted guided feature map to obtain a final output feature map; The breast segmentation model further includes a DySample upsampling module, which is used for pixel-level spatial rearrangement of the final output feature map to use dynamically generated offsets to determine the positions of upsampling points, which reduces the neglect of positional relationships and the disordered shifting of offsets; The MGSA module includes: A feature weight conversion module is used to perform Otsu binarization on the original image, segment the foreground and background, and perform Sigmoid processing on the edge feature map obtained by the Scharr operator and the reverse feature map obtained by reverse attention with the decoded prediction feature map output by the decoder, and convert them into corresponding guided feature weights; A product module is used to multiply the encoding feature map obtained by the encoding feature with the feature weight, so as to obtain a weighted guide feature map.

2. The breast segmentation model training method for breast ultrasound image lesion area recognition according to claim 1, characterized in that: The feature weight conversion module obtains the feature weight through the following formula: Guide 1i =σ(Scharr(Pred i )) Guide 2i =σ(Otsu(O ri )) Guide 3i =1-σ(Pred i ); Among them, Guide 1i Guide is the edge guide feature weight, 2i Guide is the background guide feature weight, 3i To guide the feature weights in reverse attention, Pred i To decode the predicted feature map, O ri is the original image.

3. The breast segmentation model training method for breast ultrasound image lesion area recognition according to claim 2, characterized in that: The product module obtains the weighted guided feature map through the following formula: Guide i =Cat(Guide 1i ×F ei )(Guide 2i ×F ei )(Guide 3i ×F ei ); in, Among them, Guide i Guide is a weighted guide feature map 1i Guide is the edge guide feature weight, 2i Guide is the background guide feature weight, 3i To guide the feature weights in reverse attention, Pred i To decode the predicted feature map, F ei is the encoding feature map.

4. The breast segmentation model training method for breast ultrasound image lesion area recognition according to claim 3, characterized in that: The dimension permutation of the weighted guided feature map and the capture of the inter-channel dependency relationship include: The weighted guided feature map contains multiple channels, and the spatial size of each channel is H×W. The weighted guided feature map first undergoes dimension permutation from C×H×W to W×H×C, and then captures the inter-channel dependencies through a two-layer multilayer perceptron (MLP). The first layer of MLP reduces the number of channels to (1 / 4) times the original, and then introduces nonlinearity through the ReLU activation function. The second layer of MLP restores the number of channels to the original dimension. Finally, the inverse permutation is performed to restore it to C×H×W, and the channel attention map is generated through the Sigmoid activation function. The input feature map and the channel attention map are element-wise multiplied to obtain an enhanced feature map.

5. The breast segmentation model training method for breast ultrasound image lesion area recognition according to claim 4, characterized in that: The channel shuffling operation is as follows: The enhanced feature maps are divided into (4) groups, each group contains (C / 4) channels, and the grouped feature maps are transposed to disrupt the channel order within each group. Subsequently, the disrupted feature maps are restored to their original shape (C×H×W) to obtain the shuffled feature maps.

6. The breast segmentation model training method for breast ultrasound image lesion area recognition according to claim 5, characterized in that: The spatial attention enhancement comprises: The shuffled feature map passes through a 7x7 convolution layer to reduce the number of channels to (1 / 4) times the original size. Then, it undergoes batch normalization and ReLU activation function for nonlinear transformation. Then, it passes through a second 7x7 convolution layer to restore the number of channels to the original dimension C. It then passes through a batch normalization layer and finally, a sigmoid activation function to generate a spatial attention map. The weighted guided feature map and the spatial attention map are multiplied element-wise to obtain the final output feature map.

7. A breast segmentation model training device for identifying lesion areas in breast ultrasound images, characterized in that: The breast segmentation model training device for breast ultrasound image lesion area recognition includes: A training set acquisition module, wherein the training set acquisition module is used to acquire a breast patient ultrasound image dataset, wherein the breast patient ultrasound image dataset includes at least one breast patient ultrasound image; A breast segmentation model acquisition module, wherein the breast segmentation model acquisition module is used to obtain a breast segmentation model, wherein the breast segmentation model includes an MGSA module, an encoder and a decoder, and the MGSA module is used to obtain a weighted guided feature map based on the original image; Training the breast segmentation model using the breast patient ultrasound image dataset to obtain a trained breast segmentation model; The breast segmentation model further includes a channel shuffling attention module, which is used to perform dimension permutation, inter-channel dependency capture, channel shuffling operation, and spatial attention enhancement on the weighted guided feature map to obtain a final output feature map; The breast segmentation model further includes a DySample upsampling module, which is used for pixel-level spatial rearrangement of the final output feature map to use dynamically generated offsets to determine the positions of upsampling points, which reduces the neglect of positional relationships and the disordered shifting of offsets; The MGSA module includes: A feature weight conversion module is used to perform Otsu binarization on the original image, segment the foreground and background, and perform Sigmoid processing on the edge feature map obtained by the Scharr operator and the reverse feature map obtained by reverse attention with the decoded prediction feature map output by the decoder, and convert them into corresponding guided feature weights; A product module is used to multiply the encoding feature map obtained by the encoding feature with the feature weight, so as to obtain a weighted guide feature map.

Images