CN115187621A - Automatic U-Net medical image contour extraction network integrating attention mechanism - Google Patents

Automatic U-Net medical image contour extraction network integrating attention mechanism Download PDF

Info

Publication number
CN115187621A
CN115187621A CN202111184138.8A CN202111184138A CN115187621A CN 115187621 A CN115187621 A CN 115187621A CN 202111184138 A CN202111184138 A CN 202111184138A CN 115187621 A CN115187621 A CN 115187621A
Authority
CN
China
Prior art keywords
attention
module
feature
mlp
medical image
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111184138.8A
Other languages
Chinese (zh)
Inventor
吕巨建
陈豪源
赵慧民
战荫伟
陈荣军
熊建斌
林凯瀚
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Polytechnic Normal University
Original Assignee
Guangdong Polytechnic Normal University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangdong Polytechnic Normal University filed Critical Guangdong Polytechnic Normal University
Priority to CN202111184138.8A priority Critical patent/CN115187621A/en
Publication of CN115187621A publication Critical patent/CN115187621A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/181—Segmentation; Edge detection involving edge growing; involving edge linking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20076—Probabilistic image processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses an attention mechanism-fused U-Net medical image contour automatic extraction network, which comprises an RGB image input module, wherein the output end of the RGB image input module is connected with the input end of a feature extraction module, and the feature extraction module comprises a feature coding module, a feature decoding module and an attention module; the output end of the feature extraction module is connected with the input end of the MLP, and the attention module comprises space attention and channel attention and is used for inhibiting neurons in a non-attention area; the MLP output bin is set to 2 neurons representing the probability of the foreground and background, respectively, and is followed in turn by Softmax and Marching Square. The invention integrates the attention module, improves the edge contour extraction precision, preliminarily solves the problem of fuzzy edges generated by the traditional frame, and reduces the interference of background noise, thereby basically meeting the precision requirement of the medical image contour extraction in the medical field; the process of the traditional framework is simplified, and the time and the cost for obtaining the target model are greatly saved.

Description

Automatic U-Net medical image contour extraction network integrating attention mechanism
Technical Field
The invention relates to the technical field of medical image processing, in particular to a U-Net medical image contour automatic extraction network integrating an attention mechanism.
Background
Medical images are capable of reflecting anatomical or functional tissue within a human body. Dividing a medical image into a plurality of mutually disjoint areas according to certain similarity characteristics in the medical image, namely medical image segmentation, is the most important basis in medical image analysis. Accurate, robust and rapid image segmentation is the most important step before subsequent links such as quantitative analysis, three-dimensional visualization and the like, and also lays the most fundamental foundation for important clinical applications such as image-guided surgery, radiotherapy planning and treatment evaluation.
In recent years, with the development of a deep neural network in the field of medical image processing, deep learning has become a mainstream method in a medical image segmentation task, and the practice of a plurality of researchers proves that a segmentation method based on the deep learning has a strong application potential in the field of medical image segmentation. The deep learning segmentation method is to implement segmentation of a medical image by classifying pixels. Unlike the traditional pixel or super-pixel classification method which uses manually made features, the deep learning method can automatically learn the features related to the task from the medical image and classify the pixels according to the features, thereby realizing end-to-end segmentation. Among them, U-Net is the most widely applied framework in the field of medical image segmentation at present.
The prior art is as follows: the U-Net network structure can acquire the detail information and the outline information of an image at an encoder part; then, the extracted features are passed to the decoder section through a jump connection stage; finally, feature recovery is performed by the decoder portion in conjunction with the features of the multiple scales. Due to the U-shaped structure, the U-Net can obtain a model with good effect by training with less pictures. The U-Net network can be divided into a feature extraction network and a feature fusion network, wherein the feature extraction network uses a convolution layer and a pooling layer to realize down-sampling operation, and the feature fusion network is up-sampling operation, so that the network gradually converges to a target area while the image resolution can be restored. And in the feature fusion stage, the features extracted in the same layer are fused again, so that the loss of details is avoided.
Although the medical image segmentation method based on U-Net achieves remarkable performance, obtaining accurate segmentation results is still very difficult due to the influence of noise problems, and most methods still have the problems of fuzzy edges, neglected details, manual parameter adjustment and the like. Therefore, the attention mechanism-fused U-Net medical image contour automatic extraction network is provided.
Disclosure of Invention
The invention aims to provide a U-Net medical image contour automatic extraction network integrating an attention mechanism, so as to solve the problems in the background technology.
In order to achieve the purpose, the invention provides the following technical scheme: a U-Net medical image contour automatic extraction network integrating an attention mechanism comprises an RGB image input module, wherein the output end of the RGB image input module is connected to the input end of a feature extraction module, and the feature extraction module comprises a feature coding module, a feature decoding module and an attention module;
the output end of the feature extraction module is connected to the input end of the MLP, and the attention module comprises space attention and channel attention and is used for inhibiting neurons in a non-attention area;
the MLP is used for classifying the extracted features, the output elements of the MLP are set to be 2 neurons which respectively represent the probability of the foreground and the probability of the background, and the probability of the foreground and the probability of the background are sequentially connected with Softmax and Marching Square.
The RGB image input module is used for inputting an RGB image.
The feature extraction module is used for extracting features of the RGB image, after the features of the RGB image are obtained by the feature extraction module, C-dimensional feature representation is carried out on each pixel, and local and global information is fused, so that MLP reasoning is carried out on the C-dimensional features of each pixel only once, since the task at the stage is classified in two ways, 2-dimensional information is obtained finally, the probability of a target and the probability of a non-target are represented respectively, a binary image can be obtained by comparing the target with the non-target, and finally an algorithm for extracting the contour of the binary image is adopted.
The attention module is used for removing interference information in the RGB image.
The feature encoding module uses ResNet18.
The channel attention is given by:
M c (F)＝F*Sigmoid(MLP(AvgPool(F))+MLP(MaxPool(F)))；
the spatial attention is calculated as follows, wherein
Figure BDA0003298499730000031
The method comprises the steps of splicing A and B according to channels;
Figure BDA0003298499730000032
combining the channel attention and the spatial attention, obtaining a calculation formula of the CBAM:
M(F)＝M s (M c (G (F))) + F; wherein: g (F) = Conv 2 (Conv 1 (F))。
The MLP employs a multi-layer perceptron with 3 hidden layers.
Compared with the prior art, the invention has the beneficial effects that: according to the invention, the attention module is fused in a certain way, so that the edge contour extraction precision is improved, the problem of fuzzy edges generated by a traditional frame is solved preliminarily, and the interference of background noise is reduced, thereby basically meeting the precision requirement of the medical image contour extraction in the medical field.
The invention simplifies the flow of the traditional framework, so that the time for training and deducing is relatively less, and the time and the cost for obtaining the target model are greatly saved.
The method carries out final extraction on the outline by a Marching Square algorithm, and the algorithm is simple and quick to realize and can carry out parallel processing.
Drawings
FIG. 1 is a schematic diagram of a framework structure of an automatic U-Net medical image contour extraction network according to the present invention;
FIG. 2 is a schematic diagram of a Backbone frame structure;
FIG. 3 is a schematic diagram of an MLP framework;
FIG. 4 is a diagram of an MLP call process;
FIG. 5 is a diagram of a first basic situation of Marching Squares;
FIG. 6 is a diagram of a second basic situation of Marching Squares;
FIG. 7 is a third basic case diagram of the Marching Squares;
FIG. 8 is a diagram illustrating a fourth basic case of Marching Squares;
FIG. 9 is a diagram of a fifth basic case of Marching Squares;
FIG. 10 is a diagram of a sixth basic case of Marching Squares.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
Referring to fig. 1-10, the present invention provides a technical solution: a U-Net medical image contour automatic extraction network integrating an attention mechanism comprises an RGB image input module, wherein the output end of the RGB image input module is connected to the input end of a feature extraction module, and the feature extraction module comprises a feature coding module, a feature decoding module and an attention module;
the output end of the feature extraction module is connected with the input end of the MLP, and the attention module comprises space attention and channel attention and is used for inhibiting neurons in a non-attention area;
the MLP is used for classifying the extracted features, the output elements of the MLP are set to be 2 neurons which respectively represent the probability of the foreground and the probability of the background, and the probability of the foreground and the probability of the background are sequentially connected with Softmax and Marching Square.
An RGB image is input and first subjected to feature extraction. The feature extraction module is similar to U-Net, one part is a coding network (namely a feature coding module), the other part is a decoding network (a feature decoding module), and meanwhile, the attention module is fused, and the added attention module can well remove interference information in the picture through subsequent ablation experiments.
After the characteristics are obtained, C-dimensional characteristic representation is carried out on each pixel, local and global information is fused, so that MLP reasoning is carried out on the C-dimensional characteristics of each pixel only once, and because the tasks in the stage are classified in two ways, 2-dimensional information is finally obtained and respectively represents the probability of a target and the probability of a non-target. A binary image can be obtained by comparing a target with a non-target, and finally an algorithm for extracting the contour of the binary image is adopted, wherein a Marching Squares algorithm is used for extracting the contour in an experiment.
The feature extraction module is divided into two parts, one is a feature coding module and the other is a feature decoding module. The feature coding module adopts ResNet18, has the characteristics of easy training, easy realization and relatively less parameter quantity, more importantly has a down-sampling structure, and is very suitable for rapid feature extraction.
The feature decoding module is similar to the feature decoding portion of U-Net, but adds an attention module before each upsampling. The CBAM module (i.e., attention module) is a network with attention mechanism, including spatial attention and channel attention, for suppressing neurons in non-attention areas.
The channel attention is given by:
M c (F)＝F*Sigmoid(MLP(AvgPool(F))+MLP(MaxPool(F)))；
the spatial attention is calculated as follows, where
Figure BDA0003298499730000051
The method comprises the steps of splicing A and B according to channels;
Figure BDA0003298499730000052
and combining the channel attention and the space attention to obtain a calculation formula of the CBAM.
M(F)＝M s (M c (G (F))) + F; wherein: g (F) = Conv 2 (Conv 1 (F))；
The features are usually the details of some pictures after feature coding, and the details include the details of a target and a non-target, if attention is paid at this time and the features of the non-target part are suppressed in advance, the predicted target area is more easily considered as 1 by the subsequent MLP module, which is beneficial to the subsequent MLP inference.
The feature extraction module is similar to U-Net, but unlike U-Net, the output of the module is not a probability map, but a description of the feature. Assuming that the input picture is W × H × 3, the dimension obtained after feature extraction is W × H × C, where C is the number of feature descriptions, it was found in experiments that setting C =512 makes the network more robust. For each pixel, there is a characterization of the C-dimensional vector. These features not only describe the information around the pixel, but also fuse global information, so subsequent determination only needs to infer the vector of the pixel.
The invention relates to a multi-layer perceptron (MLP) which is used for classifying extracted features. The PIFU samples and classifies the whole features, and is different from the PIFU in that the features cannot be sampled in the framework of the invention because the features are information of a two-dimensional picture and the calculated amount is not large, the calculation capability and the video memory capacity of the conventional equipment are greatly improved, all the features can be completely trained without sampling. For each pixel, there is a C-dimensional vector describing the feature information of the pixel in the global.
The frame of the invention is provided with 3 hidden layers, 512,256,128 respectively. Since the purpose of image segmentation is two-classification, the output bins are set to 2 neurons, representing the probabilities of the foreground and background, respectively, followed by Softmax and Marching Squares.
And finally comparing the probability of the foreground and the background, and setting the pixel as 1 as long as the foreground is larger than the background, so that a pure binary image can be obtained through the process.
The benefits of this are:
(1) The accuracy is improved. The conventional image segmentation is a probability map, and probability calculation is carried out on both the foreground and the background. As can be seen from ablation experiments, the introduction of MLP can improve the accuracy by about 0.3%.
(2) The threshold value is not required to be set manually. In the prior art, the extraction of the image contour usually needs to carry out a manually set threshold value on a probability map so as to obtain a pure binary image. The design of the framework avoids setting the hyper-parameter and avoids the influence of the manually set threshold on the result.
After a pure binary image is obtained, theoretically, any binary image contour extraction algorithm can extract a contour, and the Marching Squares algorithm is used in the method. Similar to Marching Cubes, marching Squares is an algorithm for extracting contour lines, a two-dimensional probability graph is given, and a curve where a threshold is located is obtained through linear interpolation according to the threshold.
Since the output of the present invention is a binary map, the same result is obtained when the Marching Squares algorithm is used, as long as the threshold is greater than 0. The reason for using the Marching Squares algorithm is that it is simple to implement and computationally inexpensive. There are 6 basic cases below for four points located in a cell, and 16 cases can be obtained by rotating the mirror image. Edges can be constructed only by judging which condition is the case, and the algorithm can run in parallel, which means that further optimization and acceleration can be realized.
The key points are as follows:
1. according to the method, on the basis of the UNet framework, an attention module is introduced, a binary mask of the medical image object is obtained by using a full convolution network, the object is segmented from the background image, the accuracy of contour extraction is improved, and features can be better provided for a contour extraction stage.
2. The method uses the multilayer perceptron with 3 hidden layers as the classifier, and carries out second correction on the mask information in the last step, so that the result is more accurate, the contour boundary is clear, and the method is favorable for extracting the contour by using a Marching Square algorithm.
3. The method uses the Marching Square as the final contour extraction algorithm, improves the performance of contour extraction compared with a neural network method, and can execute the process in parallel and in an accelerated manner.
Protection points:
1. on the basis of the traditional U-Net, an attention module is introduced in the up-sampling stage of the U-Net, a full convolution network is utilized to obtain a binary mask of a medical image object, and the object is segmented from a background image, so that the method is within the protection range of the method.
2. The invention uses a multilayer perceptron with 3 hidden layers as a classifier, and carries out secondary correction on the binary mask, so that the result is more accurate, and the contour boundary is clear, thereby better meeting the requirement of the medical field on the image contour extraction effect, and being within the protection scope of the invention.
3. The method uses the Marching Square as the final contour extraction algorithm, improves the performance of contour extraction compared with a neural network method, and can execute the process in parallel and in an accelerated manner. The application of the Marching Square algorithm in the medical contour extraction shall be within the scope of the present invention.
The prior art is as follows: the U-Net network structure can acquire the detail information and the outline information of an image at an encoder part; then, the extracted features are passed to the decoder section through a jump connection stage; finally, feature recovery is performed by the decoder portion in combination with the features of the multiple scales. Due to the U-shaped structure, the U-Net can obtain a model with good effect by training with less pictures. The U-Net network can be divided into a feature extraction network and a feature fusion network, wherein the feature extraction network uses a convolution layer and a pooling layer to realize down-sampling operation, and the feature fusion network is up-sampling operation, so that the network gradually converges to a target area while the image resolution can be restored. And in the feature fusion stage, the features extracted in the same layer are fused again, so that the loss of details is avoided.
Although the medical image segmentation method based on U-Net obtains remarkable performance, the accurate segmentation result is still difficult to obtain due to the influence of noise problems, and most methods still have the problems of fuzzy edges, neglected details, manual parameter adjustment and the like.
The prior art has the following defects:
the existing medical image contour extraction network has the following defects:
1. the detection precision is relatively low, and the phenomenon of false detection or missing detection is easy to occur.
2. Requiring extensive data set training, consuming significant costs and time.
3. The output result is edge blurred, and details are ignored.
4. Conventional networks have one or more hyper-parameters that need to be adjusted manually, and the adjustment of the hyper-parameters has a direct effect on the result.
The above disadvantages arise for the following reasons:
1. the model does not focus on certain important features and is easily interfered by noise.
2. The design of the model itself is complex, resulting in a large amount of time spent training.
3. The convolution operation on the tensor in CNN tends to make details neglected, forcing the result towards a stable state between deterministic and indeterminate.
4. The determination of the contour edge requires the artificial setting of a threshold, usually set to 0.5, but this value is not optimal and there are one or more different optimal values for different data.
In order to solve the problems of edge blurring, omission of details and the like, improve the accuracy rate and training efficiency of contour extraction and reduce the influence of human factors on hyper-parameters, in particular, aiming at the problems, the invention focuses on a network which is realized through an attention mechanism and is based on pixel judgment and does not need to use ACMs (adaptive computing machines), and can converge faster and train a good effect without a large number of data samples by combining the attention mechanism. The probability that the pixel is located in the internal area is deduced by using the network, and a binary image of the object can be directly output through certain steps without manually setting a threshold value. The main improvements of the invention are as follows:
(1) In order to solve the problem of the interference item, a U-Net module with an attention mechanism is provided, the module can effectively remove the interference, and the recall rate in the fruit fly embryo data set is improved by 6.3 percent compared with the U-Net.
(2) And (4) using MLP to express the probability that the pixel is positioned in the target, and calculating the extracted features of the U-Net again instead of directly obtaining the features of the U-Net. The accuracy rate in the fruit fly embryo data set is improved by 0.3 percent compared with that of U-Net.
(3) The manual threshold value setting greatly affects the result, and in order to avoid the manual threshold value setting, the one-hot coding method is used for solving the problem, and the method is simpler and more effective than the method for adaptively setting the threshold value.
According to the invention, the attention module is fused in a certain way, so that the edge contour extraction precision is improved, the problem that the traditional frame generates fuzzy edges is solved preliminarily, and the interference of background noise is reduced, thereby basically meeting the precision requirement of the medical image contour extraction in the medical field.
The method simplifies the flow of the traditional framework, so that the time for training and deducing is relatively less, and the time and the cost for obtaining the target model are greatly saved.
The invention carries out final extraction on the outline by a Marching Square algorithm, the algorithm is simple and quick to realize, and parallel processing can be carried out.
The technical problem solved by the invention is as follows:
1. and adding an attention module, and focusing the network on the concerned area by using a CBAM module to remove the interference.
2. The problem of manually setting the threshold is solved by using a one-hot coding method.
3. The existing model is improved and optimized, so that the training time is relatively short, and the model is lightened as much as possible.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims (7)

1. A U-Net medical image contour automatic extraction network integrating attention mechanism comprises an RGB image input module and is characterized in that: the output end of the RGB image input module is connected to the input end of the feature extraction module, and the feature extraction module comprises a feature coding module, a feature decoding module and an attention module;
the output end of the feature extraction module is connected to the input end of the MLP, and the attention module comprises space attention and channel attention and is used for inhibiting neurons in a non-attention area;
the MLP is used for classifying the extracted features, the output elements of the MLP are set to be 2 neurons which respectively represent the probability of the foreground and the probability of the background, and the probability of the foreground and the probability of the background are sequentially connected with Softmax and Marching Square.
2. The attention mechanism-fused U-Net medical image contour automatic extraction network according to claim 1, wherein: the RGB image input module is used for inputting an RGB image.
3. The attention mechanism-fused U-Net medical image contour automatic extraction network according to claim 1, wherein: the feature extraction module is used for extracting features of the RGB image, after the features of the RGB image are obtained, C-dimensional feature representation is carried out on each pixel, local and global information is fused, therefore, only one MLP reasoning needs to be carried out on the C-dimensional features of each pixel, as tasks in the stage are classified in two ways, 2-dimensional information is obtained finally, the probability of a target and the probability of a non-target are represented respectively, a binary image can be obtained by comparing the target with the non-target, and finally an algorithm for extracting binary image contours is adopted.
4. The attention mechanism-fused U-Net medical image contour automatic extraction network according to claim 1, wherein: the attention module is used for removing interference information in the RGB image.
5. The attention mechanism-fused U-Net medical image contour automatic extraction network according to claim 1, wherein: the feature encoding module uses ResNet18.
6. The attention mechanism-fused U-Net medical image contour automatic extraction network according to claim 1, wherein: the channel attention is given by:
M c (F)＝F*Sigmoid(MLP(AvgPool(F))+MLP(MaxPool(F)))；
the spatial attention is calculated as follows, wherein
Figure FDA0003298499720000021
The method comprises the following steps of (1) splicing A and B according to channels;
Figure FDA0003298499720000022
combining the channel attention and the spatial attention, a calculation formula of the CBAM is obtained:
M(F)＝M s (M c (G (F))) + F; wherein: g (F) = Conv 2 (Conv 1 (F))。
7. The attention mechanism-fused U-Net medical image contour automatic extraction network according to claim 1, wherein: the MLP employs a multi-layer perceptron with 3 hidden layers.
CN202111184138.8A 2021-10-11 2021-10-11 Automatic U-Net medical image contour extraction network integrating attention mechanism Pending CN115187621A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111184138.8A CN115187621A (en) 2021-10-11 2021-10-11 Automatic U-Net medical image contour extraction network integrating attention mechanism

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111184138.8A CN115187621A (en) 2021-10-11 2021-10-11 Automatic U-Net medical image contour extraction network integrating attention mechanism

Publications (1)

Publication Number Publication Date
CN115187621A true CN115187621A (en) 2022-10-14

Family

ID=83511597

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111184138.8A Pending CN115187621A (en) 2021-10-11 2021-10-11 Automatic U-Net medical image contour extraction network integrating attention mechanism

Country Status (1)

Country Link
CN (1) CN115187621A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116797614A (en) * 2023-03-23 2023-09-22 天津大学 CBAUnet-based double-attention rapid tongue contour extraction method and system
CN117710734A (en) * 2023-12-13 2024-03-15 北京百度网讯科技有限公司 Method, apparatus, electronic device, and medium for obtaining semantic data

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116797614A (en) * 2023-03-23 2023-09-22 天津大学 CBAUnet-based double-attention rapid tongue contour extraction method and system
CN116797614B (en) * 2023-03-23 2024-02-06 天津大学 CBAUnet-based double-attention rapid tongue contour extraction method and system
CN117710734A (en) * 2023-12-13 2024-03-15 北京百度网讯科技有限公司 Method, apparatus, electronic device, and medium for obtaining semantic data

Similar Documents

Publication Publication Date Title
Li et al. Single image dehazing via conditional generative adversarial network
Engin et al. Cycle-dehaze: Enhanced cyclegan for single image dehazing
Zhang et al. Image inpainting based on deep learning: A review
CN111950649B (en) Attention mechanism and capsule network-based low-illumination image classification method
CN112150493B (en) Semantic guidance-based screen area detection method in natural scene
Meng et al. Single-image dehazing based on two-stream convolutional neural network
CN117253154B (en) Container weak and small serial number target detection and identification method based on deep learning
Nair et al. Image mining applications for underwater environment management-A review and research agenda
CN116797488A (en) Low-illumination image enhancement method based on feature fusion and attention embedding
CN113393457B (en) Anchor-frame-free target detection method combining residual error dense block and position attention
CN115187621A (en) Automatic U-Net medical image contour extraction network integrating attention mechanism
CN113450313B (en) Image significance visualization method based on regional contrast learning
CN116935044B (en) Endoscopic polyp segmentation method with multi-scale guidance and multi-level supervision
CN113158905A (en) Pedestrian re-identification method based on attention mechanism
Wu et al. Cascaded fully convolutional DenseNet for automatic kidney segmentation in ultrasound images
Tang et al. AutoEnhancer: Transformer on U-Net architecture search for underwater image enhancement
CN115953784A (en) Laser coding character segmentation method based on residual error and feature blocking attention
CN117893823A (en) Apple maturity detection method based on Swin Transformer
Weng et al. A Lightweight Low-Light Image Enhancement Network via Channel Prior and Gamma Correction
CN113627481A (en) Multi-model combined unmanned aerial vehicle garbage classification method for smart gardens
CN112132207A (en) Target detection neural network construction method based on multi-branch feature mapping
US20230073175A1 (en) Method and system for processing image based on weighted multiple kernels
CN116129417A (en) Digital instrument reading detection method based on low-quality image
CN114565631A (en) Plant leaf fine segmentation method based on double-layer convolution network and mask refinement
Zhou et al. A lightweight object detection framework for underwater imagery with joint image restoration and color transformation

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination