CN113889234A - Medical image segmentation method based on channel mixing coding and decoding network - Google Patents

Medical image segmentation method based on channel mixing coding and decoding network Download PDF

Info

Publication number
CN113889234A
CN113889234A CN202111154112.9A CN202111154112A CN113889234A CN 113889234 A CN113889234 A CN 113889234A CN 202111154112 A CN202111154112 A CN 202111154112A CN 113889234 A CN113889234 A CN 113889234A
Authority
CN
China
Prior art keywords
medical image
characteristic diagram
channel
feature map
feature
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
CN202111154112.9A
Other languages
Chinese (zh)
Inventor
田辉
刘其开
郭玉刚
张志翔
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.)
Hefei High Dimensional Data Technology Co ltd
Original Assignee
Hefei High Dimensional Data Technology Co ltd
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 Hefei High Dimensional Data Technology Co ltd filed Critical Hefei High Dimensional Data Technology Co ltd
Priority to CN202111154112.9A priority Critical patent/CN113889234A/en
Publication of CN113889234A publication Critical patent/CN113889234A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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
    • G06T7/10—Segmentation; Edge detection

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Epidemiology (AREA)
  • Medical Informatics (AREA)
  • Primary Health Care (AREA)
  • Public Health (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

The invention particularly relates to a medical image segmentation method based on a channel mixing coding and decoding network, which comprises the following steps: acquiring a medical image to be segmented; importing the medical image to be segmented into a trained network model for recognition to obtain a segmented medical image; the trained network model infrastructure is a symmetrical U-Net network structure, and the characteristic diagrams of the decoder part in the U-Net network structure are obtained by respectively processing and fusing the characteristic diagrams with the same size in an encoder through a self-attention module and a maximum up-sampling index matrix. According to the medical image segmentation method provided by the invention, most segmentation algorithms use skip connection of U-Net for reference to fuse information under different scales, so that the aim of information transmission is guided; different from the traditional algorithm, the method provides a channel-mixed self-attention module to replace a skip-connect structure and maximum index upsampling to replace a transposed convolution, so as to realize the feature upsampling of the decoding network.

Description

Medical image segmentation method based on channel mixing coding and decoding network
Technical Field
The invention relates to the technical field of computer image recognition, in particular to a medical image segmentation method based on a channel mixing coding and decoding network.
Background
Currently, computer vision technology is applied to a plurality of scenes, including the fields of image classification, target detection, three-dimensional reconstruction, semantic segmentation, and the like. With the rapid development of internet communication, the competitiveness of intelligent products requires a technical breakthrough of higher semantic scene understanding. Therefore, semantic segmentation is used as a core problem of computer vision, and can help more and more products to automatically and efficiently understand related knowledge or semantics in images or videos, so that an intelligent target is achieved, manual interactive operation is reduced, and comfort of customers is improved. These products are currently used in a wide variety of applications in the fields of automotive driving, human-computer interaction, computational photography, image search engines, augmented reality, and the like.
The semantic segmentation problem in computer vision is essentially a process that progresses from coarse to refined reasoning. Going back to the classification problem, i.e. roughly predicting the object class in the input sample, is followed by the location and detection of the target object, which not only predicts the class of the object, but also gives additional information about the spatial location of each class, such as the center point or the border of the object area. On the basis, semantic segmentation can be understood as fine-grained prediction in the detection field, a test image is input into a segmentation network, the size of a predicted heat map is consistent with that of an input image, the number of channels is equal to the number of classes, the probabilities that all spatial positions belong to all the classes are represented respectively, and classification can be carried out on a pixel-by-pixel basis.
The full convolution network FCN becomes a base for applying a deep learning technology to a semantic segmentation problem, can accept an input image with any size, and performs up-sampling decoding on a feature map (feature map) of the last convolution of a coding network through a plurality of deconvolution layers to restore the feature map to the same size of the input image, so that a prediction can be generated for each pixel, and spatial information in the original input image is kept. Then, on the basis of the FCN, a plurality of semantic segmentation models are derived, such as a symmetric network U-net with jump connection between encoding and decoding, a DeepLab series network introducing expansion volume and post-processing optimization by using a conditional random field CRF, and a ParseNet combining context information for feature fusion.
The medical image segmentation is a complex and key step in the field of medical image processing and analysis, and aims to segment parts with certain special meanings in a medical image, extract relevant features, provide reliable basis for clinical diagnosis and pathological research and assist doctors in making more accurate diagnosis. However, it is still a difficult task to automatically segment the target from the medical image by using some common algorithms, and the segmentation accuracy is not high due to the fact that the medical image has high complexity and lacks of simple linear features, and partial volume effect, gray level non-uniformity, artifacts, different gray values between soft tissues, and the like.
Disclosure of Invention
The invention aims to provide a medical image segmentation method based on a channel mixing coding and decoding network, which can better realize the segmentation of complex images.
In order to realize the purpose, the invention adopts the technical scheme that: a medical image segmentation method based on a channel mixed coding and decoding network comprises the following steps: acquiring a medical image to be segmented; importing the medical image to be segmented into a trained network model for recognition to obtain a segmented medical image; the trained network model infrastructure is a symmetrical U-Net network structure, and the characteristic diagrams of the decoder part in the U-Net network structure are obtained by respectively processing and fusing the characteristic diagrams with the same size in an encoder through a self-attention module and a maximum up-sampling index matrix.
Compared with the prior art, the invention has the following technical effects: according to the medical image segmentation method provided by the invention, most segmentation algorithms use skip connection of U-Net for reference to fuse information under different scales, so that the aim of information transmission is guided; different from the traditional algorithm, the method provides a channel-mixed self-attention module to replace a skip-connect structure and the maximum index upsampling to replace the transposition convolution, so as to realize the characteristic upsampling of the decoding network; in addition, a combined loss strategy of the segmented training is also provided to achieve the balance of the training speed and the training precision.
Drawings
FIG. 1 is a block diagram of the architecture of a network model in the present invention;
FIG. 2 is a detailed block diagram of a network model in the present invention;
FIG. 3 is a detailed block diagram of the self-attention module of the present invention;
fig. 4 is a schematic diagram of a maximum upsampling index matrix concatenation.
Detailed Description
The present invention will be described in further detail with reference to fig. 1 to 4.
Referring to fig. 1, a method for segmenting a medical image based on a channel-mixed codec network includes the following steps: acquiring a medical image to be segmented; importing the medical image to be segmented into a trained network model for recognition to obtain a segmented medical image; the trained network model infrastructure is a symmetrical U-Net network structure, and the characteristic diagrams of the decoder part in the U-Net network structure are obtained by respectively processing and fusing the characteristic diagrams with the same size in an encoder through a self-attention module and a maximum up-sampling index matrix. According to the medical image segmentation method provided by the invention, most segmentation algorithms use skip connection of U-Net for reference to fuse information under different scales, so that the aim of information transmission is guided; different from the traditional algorithm, the method provides a channel-mixed self-attention module to replace a skip-connect structure and maximum index upsampling to replace a transposed convolution, so as to realize the feature upsampling of the decoding network. The self-attention module can encode the global feature information of a large-range area and then add the global feature information into the local feature information, so that the local feature of the feature map has information dependent on the global spatial feature, and the representation capability of the module is further enhanced. The maximum upsampling index matrix is used for storing position information of the pooled points in the symmetric encoding stage, and in the decoding stage, for the upsampling operation, which region of the previous 2x2 each pooled 1x1 feature point comes from is determined according to the index matrix of the symmetric encoding record. Through the processing of the self-attention module and the maximum up-sampling index matrix, more information can be fused when the feature map is subjected to image semantic reduction, and the final segmentation precision is improved.
There are many setting methods for the U-Net network structure based on the self-attention module and the maximum upsampling index matrix, and the scheme shown in fig. 2 is preferably adopted in the present invention, because the network structure in the present invention is complicated and is not convenient to be directly expressed from the structure, the network structure is presented here by describing the specific steps of image processing. Specifically, the U-Net network structure processes an input image into a segmented medical image according to the following steps: s100, performing convolution operation on the input image to obtain a feature map a1The number of channels of the input image, height and width, is marked as C, H and W; s200, matching feature graphs aiPerforming convolution and pooling operation to obtain a feature map ai+1Where i ∈ {1,2, …, n-1}, i.e.: for characteristic diagram a1Performing convolution and pooling operation to obtain a feature map a2For feature map a2Performing convolution and pooling operation to obtain a feature map a3And so on, finally obtaining the characteristic diagram an(ii) a S300, utilizing the self-attention module to pair the feature map aiProcessing is carried out, and the maximum upsampling index matrix is utilized to carry out processing on the characteristic diagram bi+1Processing, and fusing the processing results of the two to obtain a characteristic diagram bi(ii) a S400, pair feature graph b1Performing convolution operation to obtain a characteristic diagram c; s500, performing softmax operation on the feature map c to obtain a segmented medical image; in the above steps, the input image and the feature map c have the same width and height, the number of channels of the feature map c is K, K is the number of categories, and the feature value in each category represents the probability that each pixel position is in the category. In the step S500, the number K of channels in the feature map c is the number of categories after the medical image is segmented; the segmented medical image is a heat map, and each K category in the heat map is displayed in different colors. Characteristic diagram aiAnd biThe sizes are completely the same, namely the number, height and width of the two channels are all equal. Through these steps, the input image can be conveniently processed into a classified heat map.
The description is made here specifically as follows: the general U-Net network model comprises an encoder and a decoder, wherein the encoder performs high-level semantic feature extraction on a feature map, the decoder performs image semantic restoration on the feature map, the encoder and the decoder are connected through a bottommost feature map, and the bottommost feature map can be divided into the encoder and the decoder, and can also exist independently as the bottommost feature map. In the present invention, for convenience of description, the characteristic diagram anAnd characteristic diagram bnAnd the bottommost characteristic diagram refers to the same thing, namely the characteristic diagram of C4H 3W 3 in FIG. 2.
In the conventional algorithm model, the up-sampling is generally realized by transpose convolution and the feature map in the encoder is directly copied into the decoder, and the improvement is aimed at. There are many specific schemes capable of implementing the self-attention module and the maximum upsampling index matrix processing, and in the present invention, it is preferable that in step S200, the feature map aiGenerating a maximum upsampling index matrix d when performing pooling operationsiThe matrix is shown in fig. 4, which expresses pooled point location information for the encoding phase. Step S300 includes the steps of: s310, characteristic diagram aiObtaining a characteristic diagram b after the self-attention module processingi,1(ii) a S320, matching feature graphs bi+1Performing convolution operation to obtain characteristic diagram
Figure 100002_DEST_PATH_IMAGE001
(ii) a S330, characteristic diagram
Figure 575222DEST_PATH_IMAGE001
Indexing the matrix d according to the maximum upsamplingjMaximum pooling upsampling to obtain a feature map bi,2(ii) a S340, converting the characteristic diagram bi,1And a characteristic diagram bi,2Performing channel fusion to obtain a feature map bi(ii) a In the above steps, the characteristic diagram ai、bi,1、bi,2、biThe sizes are completely the same; the channel fusion is as follows: from the characteristic map bi,1Selecting partial channels, from feature map bi,2Selecting another part of channels to jointly form a characteristic diagram biIn the present invention, preferably, in step S340, the feature maps b are respectively selected fromi,1And a characteristic diagram bi,2The number of channels of 1/2 constitutes a feature map bi. Through the steps, the image voice is restored, and the finally obtained classification heat map is more accurate.
Referring to fig. 3, further, the step S310 includes the following steps: s311, comparing the feature map aiCarrying out convolution operation on partial channels to respectively obtain a characteristic diagram f and a characteristic diagram g; s312, matching feature graphs aiPerforming convolution operation to obtain a characteristic graph h; s313, processing the feature maps f and g into two-dimensional matrixes with the channel number x (height and width) according to the channel number, height and width of the feature maps; s314, after the two-dimensional matrix corresponding to the characteristic diagram f is rotated, multiplying the two-dimensional matrix corresponding to the characteristic diagram g to obtain an attention layer matrix; s315, multiplying the characteristic diagram h with the attention layer matrix to obtain a characteristic diagram bi,1. Preferably, in step S311, the "part" of the partial channel is 1/4 or 1/8, that is: channel number of characteristic diagram f and characteristic diagram g and characteristic diagram aiThe ratio of the number of channels of (a) is 1/4 or 1/8. 1/8 used in this example is described in detail below in conjunction with FIG. 3.
Suppose a feature map a to be processediC × H × W, the "part" of the partial channel is equal to 1/8, and after the convolution operation, the sizes of the feature maps f and g are both (C/8) × 0H × 1W, and at this time, the two feature maps are converted into a two-dimensional matrix, that is, a matrix of (C/8) × (H × W) size, that is, a matrix having C/8 columns and H × W elements in each row. Then transposing a matrix corresponding to the characteristic diagram f to obtain a matrix (H multiplied by W) x (C/8), and multiplying the matrix and the matrix corresponding to the characteristic diagram g to obtain a two-dimensional matrix (H multiplied by W) x (H multiplied by W) in size, wherein the matrix is an attention layer matrix; finally, multiplying the characteristic diagram h by the attention layer matrix to obtain the characteristic diagram aiFeature map b of the same sizei,1. Through the steps, the global feature information of a large-range area can be coded and then added into the local feature information, so that the local feature of the feature map has information depending on the global spatial feature.
For medical image segmentation, cross entropy loss or dice loss is used as a common loss function. In the present invention, the weighting of the two is adopted, and the weighting and the time coefficient
Figure 143607DEST_PATH_IMAGE002
Correlation, in particular, loss function in the training of the U-Net network
Figure DEST_PATH_IMAGE003
The following were used:
Figure 100002_DEST_PATH_IMAGE005
in the formula (I), the compound is shown in the specification,
Figure 766349DEST_PATH_IMAGE006
representing the prediction probability that the ith sample belongs to the jth class,
Figure 100002_DEST_PATH_IMAGE007
indicates the probability that the ith sample belongs to the jth class label, N indicates the number of samples, K indicates the number of classes,
Figure 101385DEST_PATH_IMAGE008
represents a smoothing factor;
Figure 188289DEST_PATH_IMAGE002
is a time coefficient used to change the weighting of the two losses. The dice loss has a good optimization effect on the problem of unbalanced sample categories, but when the values of p and q are very small, the calculated gradient value may be very large, which can cause instability of training, so that a time coefficient is introduced in the scheme
Figure 802941DEST_PATH_IMAGE002
In the early stage of training,
Figure 772034DEST_PATH_IMAGE002
close to 1, loss function loss of
Figure 100002_DEST_PATH_IMAGE009
Mainly, as the training progresses,
Figure 582865DEST_PATH_IMAGE002
gradual decrease, 1-
Figure 852172DEST_PATH_IMAGE010
Is gradually increased so that
Figure 700042DEST_PATH_IMAGE009
The weight is reduced and the weight is lowered,
Figure 100002_DEST_PATH_IMAGE011
the weights are increased, and in the late training phase,
Figure 828535DEST_PATH_IMAGE002
close to 0, loss function loss of
Figure 787264DEST_PATH_IMAGE011
Mainly comprises the following steps.
Time coefficient
Figure 770132DEST_PATH_IMAGE002
The calculation modes are various, and only the numerical value of the training early stage is required to be smaller than and close to 1, and the numerical value of the training later stage is required to be larger than and close to 0. Preferably, in the present invention, the calculation formula of the time coefficient is as follows:
Figure 585642DEST_PATH_IMAGE012
t represents the ratio of the current iteration number to the lost switching number T,
Figure DEST_PATH_IMAGE013
and T is a defined hyperparameter, said
Figure 404693DEST_PATH_IMAGE013
The value ranges from 5 to 10. Time coefficient thus set
Figure 167113DEST_PATH_IMAGE002
Two determined loss weights are more reasonable.
After the network model is constructed according to the previous steps, it needs to be trained. During training, the method can be carried out according to the following steps:
1. preparing a public data set, and determining the input size of the data set;
2. building training network, determining each network parameter and hyper-parameter during training, the hyper-parameter mentioned above
Figure 879854DEST_PATH_IMAGE013
And T, and also comprises parameters of the conventional U-Net network during training, such as initial learning rate, total epoch and the like, wherein the learning rate is reduced after iteration for a certain number of times
Figure 993828DEST_PATH_IMAGE002
Also, as the iteration progresses, its value gradually decreases;
3. inputting a sample set, and performing batch sampling training;
4. and finally, verifying and testing the model by using a verification set, and identifying the picture to be detected after the tested network model is stored.

Claims (10)

1. A medical image segmentation method based on a channel mixed coding and decoding network is characterized in that: the method comprises the following steps:
acquiring a medical image to be segmented;
importing the medical image to be segmented into a trained network model for recognition to obtain a segmented medical image;
the trained network model infrastructure is a symmetrical U-Net network structure, and the characteristic diagrams of the decoder part in the U-Net network structure are obtained by respectively processing and fusing the characteristic diagrams with the same size in an encoder through a self-attention module and a maximum up-sampling index matrix.
2. The medical image segmentation method based on the channel mixing coding and decoding network as claimed in claim 1, characterized in that: the U-Net network structure processes an input image into a segmented medical image according to the following steps:
s100, performing convolution operation on the input image to obtain a feature map a1；
S200, matching feature graphs aiPerforming convolution and pooling operation to obtain a feature map ai+1Where i ∈ {1,2, …, n-1}, anNamely the bottommost characteristic diagram bn；
S300, utilizing the self-attention module to pair the feature map aiProcessing is carried out, and the maximum upsampling index matrix is utilized to carry out processing on the characteristic diagram bi+1Processing, and fusing the processing results of the two to obtain a characteristic diagram bi；
S400, pair feature graph b1Performing convolution operation to obtain a characteristic diagram c;
s500, performing softmax operation on the feature map c to obtain a segmented medical image;
in the above steps, the input image has the same width and height as the feature map c, and the feature map aiAnd biThe dimensions are identical.
3. The medical image segmentation method based on the channel mixing coding and decoding network as claimed in claim 1, characterized in that: in the step S200, the feature map aiGenerating a maximum upsampling index matrix d when performing pooling operationsiStep S300 includes the following steps:
s310, characteristic diagram aiObtaining a characteristic diagram b after the self-attention module processingi,1；
S320, matching feature graphs bi+1Performing convolution operation to obtain characteristic diagram
Figure DEST_PATH_IMAGE001
；
S330, characteristic diagram
Figure 173478DEST_PATH_IMAGE001
Indexing the matrix d according to the maximum upsamplingjMaximum pooling upsampling to obtain a feature map bi,2；
S340, converting the characteristic diagram bi,1And a characteristic diagram bi,2Performing channel fusion to obtain a feature map bi；
In the above steps, the characteristic diagram ai、bi,1、bi,2、biThe sizes are completely the same; the channel fusion is as follows: from the characteristic map bi,1Selecting partial channels, from feature map bi,2Selecting another part of channels to jointly form a characteristic diagram bi。
4. The method for segmenting medical images based on the channel mixing coding and decoding network as claimed in claim 3, characterized in that: the step S310 includes the following steps:
s311, comparing the feature map aiCarrying out convolution operation on partial channels to respectively obtain a characteristic diagram f and a characteristic diagram g;
s312, matching feature graphs aiPerforming convolution operation to obtain a characteristic graph h;
s313, processing the feature maps f and g into two-dimensional matrixes with the channel number x (height and width) according to the channel number, height and width of the feature maps;
s314, after the two-dimensional matrix corresponding to the characteristic diagram f is rotated, multiplying the two-dimensional matrix corresponding to the characteristic diagram g to obtain an attention layer matrix;
s315, multiplying the characteristic diagram h with the attention layer matrix to obtain a characteristic diagram bi,1。
5. The method for segmenting medical images based on the channel mixing coding and decoding network as claimed in claim 3, characterized in that: in step S340, the feature maps b are respectively selectedi,1And a characteristic diagram bi,2The number of channels of 1/2 constitutes a feature map bi。
6. The medical image segmentation method based on the channel mixing coding and decoding network as claimed in claim 2, characterized in that: in the step S500, the number K of channels in the feature map c is the number of categories after the medical image is segmented; the segmented medical image is a heat map, and each K category in the heat map is displayed in different colors.
7. The method for segmenting medical images based on the channel mixing coding and decoding network as claimed in claim 4, characterized in that: in the step S311, the "part" of the partial channel is 1/4 or 1/8.
8. The method for segmenting medical images based on channel-mixed codec networks according to any of claims 1 to 7, characterized in that: loss function in the training of the U-Net network
Figure 183022DEST_PATH_IMAGE002
The following were used:
Figure 623231DEST_PATH_IMAGE004
in the formula (I), the compound is shown in the specification,
Figure DEST_PATH_IMAGE005
representing the prediction probability that the ith sample belongs to the jth class,
Figure 339821DEST_PATH_IMAGE006
indicates the probability that the ith sample belongs to the jth class label, N indicates the number of samples, K indicates the number of classes,
Figure DEST_PATH_IMAGE007
represents a smoothing factor;
Figure 599901DEST_PATH_IMAGE008
is a time coefficient used to change the weighting of the two losses.
9. The method of claim 8 for medical image segmentation based on channel-mixing codec networkThe method is characterized in that: the calculation formula of the time coefficient is as follows:
Figure DEST_PATH_IMAGE009
t represents the ratio of the current iteration number to the lost switching number T,
Figure 881978DEST_PATH_IMAGE010
and T is the determined hyperparameter.
10. The medical image segmentation method based on the channel mixing coding and decoding network as claimed in claim 9, characterized in that: said
Figure DEST_PATH_IMAGE011
The value ranges from 5 to 10.
CN202111154112.9A 2021-09-29 2021-09-29 Medical image segmentation method based on channel mixing coding and decoding network Pending CN113889234A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111154112.9A CN113889234A (en) 2021-09-29 2021-09-29 Medical image segmentation method based on channel mixing coding and decoding network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111154112.9A CN113889234A (en) 2021-09-29 2021-09-29 Medical image segmentation method based on channel mixing coding and decoding network

Publications (1)

Publication Number Publication Date
CN113889234A true CN113889234A (en) 2022-01-04

Family

ID=79008408

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111154112.9A Pending CN113889234A (en) 2021-09-29 2021-09-29 Medical image segmentation method based on channel mixing coding and decoding network

Country Status (1)

Country Link
CN (1) CN113889234A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114419318A (en) * 2022-01-18 2022-04-29 北京工业大学 Medical image segmentation method based on deep learning
CN114612479A (en) * 2022-02-09 2022-06-10 苏州大学 Medical image segmentation method based on global and local feature reconstruction network
CN115731243A (en) * 2022-11-29 2023-03-03 北京长木谷医疗科技有限公司 Spine image segmentation method and device based on artificial intelligence and attention mechanism

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114419318A (en) * 2022-01-18 2022-04-29 北京工业大学 Medical image segmentation method based on deep learning
CN114612479A (en) * 2022-02-09 2022-06-10 苏州大学 Medical image segmentation method based on global and local feature reconstruction network
CN115731243A (en) * 2022-11-29 2023-03-03 北京长木谷医疗科技有限公司 Spine image segmentation method and device based on artificial intelligence and attention mechanism
CN115731243B (en) * 2022-11-29 2024-02-09 北京长木谷医疗科技股份有限公司 Spine image segmentation method and device based on artificial intelligence and attention mechanism

Similar Documents

Publication Publication Date Title
CN114529825B (en) Target detection model, method and application for fire fighting access occupied target detection
CN113889234A (en) Medical image segmentation method based on channel mixing coding and decoding network
CN111738363B (en) Alzheimer disease classification method based on improved 3D CNN network
CN111210435A (en) Image semantic segmentation method based on local and global feature enhancement module
CN110246148B (en) Multi-modal significance detection method for depth information fusion and attention learning
CN112396607A (en) Streetscape image semantic segmentation method for deformable convolution fusion enhancement
CN114943963A (en) Remote sensing image cloud and cloud shadow segmentation method based on double-branch fusion network
CN110929736A (en) Multi-feature cascade RGB-D significance target detection method
CN110490082A (en) A kind of road scene semantic segmentation method of effective integration neural network characteristics
CN112132834B (en) Ventricular image segmentation method, ventricular image segmentation system, ventricular image segmentation device and storage medium
CN111798469A (en) Digital image small data set semantic segmentation method based on deep convolutional neural network
CN111401436A (en) Streetscape image segmentation method fusing network and two-channel attention mechanism
CN116309648A (en) Medical image segmentation model construction method based on multi-attention fusion
CN113192073A (en) Clothing semantic segmentation method based on cross fusion network
CN114549574A (en) Interactive video matting system based on mask propagation network
CN116469100A (en) Dual-band image semantic segmentation method based on Transformer
CN113111906B (en) Method for generating confrontation network model based on condition of single pair image training
CN117351363A (en) Remote sensing image building extraction method based on transducer
CN116703885A (en) Swin transducer-based surface defect detection method and system
CN116205962B (en) Monocular depth estimation method and system based on complete context information
CN114049314A (en) Medical image segmentation method based on feature rearrangement and gated axial attention
CN117095287A (en) Remote sensing image change detection method based on space-time interaction transducer model
Yang et al. Xception-based general forensic method on small-size images
CN116596966A (en) Segmentation and tracking method based on attention and feature fusion
CN114463187B (en) Image semantic segmentation method and system based on aggregation edge features

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
CB02 Change of applicant information
CB02 Change of applicant information

Address after: 230088 21 / F, building A1, phase I, Zhongan chuanggu Science Park, No. 900, Wangjiang West Road, high tech Zone, Hefei, Anhui

Applicant after: HEFEI HIGH DIMENSIONAL DATA TECHNOLOGY Co.,Ltd.

Address before: 230088 Block C, building J2, innovation industrial park, 2800 innovation Avenue, high tech Zone, Hefei City, Anhui Province

Applicant before: HEFEI HIGH DIMENSIONAL DATA TECHNOLOGY Co.,Ltd.