CN117830324A - 3D medical image segmentation method based on multi-dimensional and global local combination - Google Patents
3D medical image segmentation method based on multi-dimensional and global local combination Download PDFInfo
- Publication number
- CN117830324A CN117830324A CN202311852915.0A CN202311852915A CN117830324A CN 117830324 A CN117830324 A CN 117830324A CN 202311852915 A CN202311852915 A CN 202311852915A CN 117830324 A CN117830324 A CN 117830324A
- Authority
- CN
- China
- Prior art keywords
- dimensional
- segmentation
- medical image
- net
- loss function
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000003709 image segmentation Methods 0.000 title claims abstract description 32
- 230000011218 segmentation Effects 0.000 claims abstract description 40
- 230000006870 function Effects 0.000 claims abstract description 28
- 230000002457 bidirectional effect Effects 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 9
- 230000004927 fusion Effects 0.000 claims description 7
- 239000003550 marker Substances 0.000 claims description 6
- 230000001105 regulatory effect Effects 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 230000000903 blocking effect Effects 0.000 claims description 3
- 230000010354 integration Effects 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 230000004807 localization Effects 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
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
-
- 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
- G06N3/096—Transfer learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a 3D medical image segmentation method based on multi-dimensional and global local combination, which belongs to the technical field of medical image processing and comprises the following steps: step 1, processing an unlabeled image and a marked image through two parallel tasks to obtain a corresponding loss function; step 2, summing the loss functions obtained in the step 1 to obtain a total loss function; step 3, updating a student model of the 3D medical image segmentation network by using random gradient descent of the total loss function; and 4, repeating the steps 1 to 3 until the total loss function is not reduced, and optimizing the 3D medical image segmentation. The invention provides a 3D medical image segmentation method based on multi-dimensional and global local combination, which has the advantages of accurate segmentation, fine segmentation and reliable prediction.
Description
Technical Field
The invention belongs to the technical field of medical image processing, and particularly relates to a 3D medical image segmentation method based on multi-dimensional and global local combination.
Background
With the rapid development of deep learning, many image segmentation methods have shown their effectiveness in facilitating quantitative analysis of medical images. Fully supervised image segmentation requires large amounts of well annotated data to ensure satisfactory performance, while manually annotating the data is very expensive and time consuming. The semi-supervised segmentation achieves the purpose of accurate segmentation by using a small amount of marked data and a large amount of unmarked data, and has important application value in the aspects of medical image processing, disease diagnosis, treatment scheme formulation and the like.
The existing semi-supervised medical image segmentation method reduces a great deal of cost and manpower compared with the full-supervised medical image segmentation method, but has some defects. 1) In the three-dimensional field, the currently known image segmentation network mostly uses a complete three-dimensional image as an input of the network, but feature extraction in a three-dimensional space can only capture finer global feature information, and local feature differences are relatively weak in the three-dimensional space, so that the accuracy of three-dimensional segmentation is not high. 2) Segmentation based on the entire image may be not fine for fine branches and edge segmentation. 3) Semi-supervised segmentation is commonly used to teachers and students models, but predictions of unlabeled data by the teacher network may be unreliable and noisy. Therefore, there is a need for a 3D medical image segmentation method based on multi-dimensional and global local integration, which solves the problems of the segmentation method.
Disclosure of Invention
The invention aims to provide a 3D medical image segmentation method based on multi-dimensional and global local combination, which solves the problems of low accuracy of three-dimensional segmentation, imprecision of segmentation of fine branches and edges, unreliable prediction and noise existing in the prior art.
In order to achieve the above object, the present invention provides a 3D medical image segmentation method based on multi-dimensional and global local association, comprising the steps of:
step 1, processing an unlabeled image and a marked image through two parallel tasks to obtain a corresponding loss function;
step 2, summing the loss functions obtained in the step 1 to obtain a total loss function;
step 3, updating a student model of the 3D medical image segmentation network by using random gradient descent of the total loss function;
and 4, repeating the steps 1 to 3 until the total loss function is not reduced, and optimizing the 3D medical image segmentation.
Preferably, the two parallel tasks in the step 1 comprise a multidimensional joint segmentation task and an image block contrast learning task.
Preferably, the specific process of the multi-dimensional joint segmentation task is as follows:
a1, two unlabeled imagesAnd->Inputting a 3D U-Net teacher model and obtaining a reliable pseudo tag by a multi-level confidence coefficient module>And->The pseudo tag and the real tag are copied and stuck in two directions to obtain a mixed tag Y out And Y in ；
A2, two marker imagesAnd->And two unlabeled images->And->X is obtained through bidirectional copy and paste out And X in Input 3D U-Net student model, output Q out And Q in ；
A3, X is out And X in Input two-dimensional consistency learning module to obtainAnd->
A4、And Q is equal to out Obtaining->And Q is equal to in Obtained through a fusion module
A5、Respectively with Y out And Y in Comparing and calculating loss to obtain L out And L in Two segmentation loss functions.
Preferably, the specific process of the image block contrast learning task is as follows:
b1, for two marker imagesAnd->And two unlabeled images->And->X obtained by bidirectional copy and paste out And X in Performing blocking treatment;
b2, inputting the obtained small blocks into a 3D U-Net teacher model and a 3D U-Net student model respectively to obtain output;
b3, calculating the block contrast learning loss functionAnd->The outputs of the same block through the 3D U-Net teacher model and the 3D U-Net student model are directly opposite, and vice versa.
Preferably, the calculation expression of the total loss function in step 2 is as follows:
wherein L is all Represents the total loss function, L in And L out The segmentation loss between the representation and the label,and->Representing block contrast learning penalty, M is a mask, < ->And->Representing a fusion result of the three-dimensional segmentation result and the two-dimensional segmentation result, < >>Y out And Y in Representing a mixed label obtained after the two-way copy pasting of a pseudo label and a real label, L seg Is a linear combination of the Dice loss and the cross entropy loss, and the Dice loss calculation formula is +.>Wherein |X| and |Y| respectively represent the real label and the prediction mask of the segmentation, N represents the number of samples, m represents the number of image blocks, d is the embedding distance, and the calculation formula is X represents out Results of each pixel of (a) output by the student model, a->X represents out Results of the respective pixels output by the teacher model, and (2)>X represents in Results of each pixel of (a) output by the student model, a->X represents in Results of each pixel of (3) output by the teacher model.
Preferably, two unlabeled images in A1And->Inputting a 3D U-Net teacher model and obtaining a reliable pseudo tag by a multi-level confidence coefficient module>And->The specific process of (2) is as follows: the 3D U-Net teacher model comprises an encoder and a decoder, wherein the decoder comprises a plurality of layers of networks, the output of each layer of the three layers is regulated to be the same by using bilinear interpolation, and the channel pooling is used for regulating the channel number of each layer of output to be equal; and obtaining a predicted value by utilizing the Monte Carlo Dropout, wherein the entropy has a fixed numerical range, measuring the multi-scale uncertainty u by utilizing the predicted entropy, and obtaining a reliable pseudo tag by utilizing u as an adaptive weight and correspondingly if the accuracy of pseudo tag prediction is high, wherein the occupied weight is large.
Preferably, the two-dimensional consistency learning module in the A3 performs dimension reduction operation by acquiring two-dimensional slices of the three-dimensional input image frame by frame, and takes the two-dimensional subgraph as input of a two-dimensional segmentation network, wherein the two-dimensional segmentation network obtains finer local differences from slice layers.
Therefore, the invention adopts the 3D medical image segmentation method based on the combination of the multidimensional and global local, and in order to encourage unlabeled data to learn comprehensive semantics from labeled data, the scheme designs a bidirectional copy-paste model based on a student model and a teacher model, and the distribution alignment is further promoted through a consistent learning strategy of the labeled data and the unlabeled data, so that the segmentation performance is improved. Aiming at the problem that local feature difference is difficult to extract in a three-dimensional space, the scheme designs a multi-dimensional combined network, and more abundant feature information is extracted from two dimensions and three dimensions. Aiming at the problem that segmentation based on the whole image is possibly not fine, the scheme designs a global local joint strategy of image block contrast learning, and utilizes the contrast learning of the image block to extract finer local feature details. Aiming at the problem that the pseudo tag output by the teacher model is possibly unreliable, the scheme designs a multi-level confidence coefficient module to select the high-certainty pseudo tag and the real tag to be copied and pasted in two directions to form a mixed tag, and the mixed tag is used for supervising the output of the input image through the student model.
The technical scheme of the invention is further described in detail through the drawings and the embodiments.
Drawings
FIG. 1 is a general flow chart of a 3D medical image segmentation method based on multi-dimensional and global local federation of the present invention;
FIG. 2 is a block diagram of a two-dimensional consistency learning module of the present invention;
FIG. 3 is a block diagram of a fusion module of the present invention;
FIG. 4 is a block diagram of image block contrast learning of the present invention;
FIG. 5 is a block diagram of a multi-level confidence module of the present invention.
Detailed Description
The following detailed description of the embodiments of the invention, provided in the accompanying drawings, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1-5, a 3D medical image segmentation method based on multi-dimensional and global local association includes the steps of:
step 1, processing an unlabeled image and a marked image through two parallel tasks to obtain a corresponding loss function, wherein the two parallel tasks comprise a multidimensional joint segmentation task and an image block comparison learning task;
the specific process of the multidimensional joint segmentation task is as follows:
a1, two unlabeled imagesAnd->Inputting a 3D U-Net teacher model and obtaining a reliable pseudo tag by a multi-level confidence coefficient module>And->The pseudo tag and the real tag are copied and stuck in two directions to obtain a mixed tag Y out And Y in The method comprises the steps of carrying out a first treatment on the surface of the Two unlabeled images->And->Inputting a 3D U-Net teacher model and obtaining a reliable pseudo tag by a multi-level confidence coefficient module>Andthe specific process of (2) is as follows: the 3D U-Net teacher model comprises an encoder and a decoder, wherein the decoder comprises a plurality of layers of networks, the output of each layer of the three layers is regulated to be the same by using bilinear interpolation, and the channel pooling is used for regulating the channel number of each layer of output to be equal; and obtaining a predicted value by utilizing the Monte Carlo Dropout, wherein the entropy has a fixed numerical range, measuring the multi-scale uncertainty u by utilizing the predicted entropy, and obtaining a reliable pseudo tag by utilizing u as an adaptive weight and correspondingly if the accuracy of pseudo tag prediction is high, wherein the occupied weight is large.
A2, two marker imagesAnd->And two unlabeled images->And->X is obtained through bidirectional copy and paste out And X in Input 3D U-Net student model, output Q out And Q in ；
A3, X is out And X in Input two-dimensional consistency learning module to obtainAnd->The two-dimensional consistency learning module performs dimension reduction operation by acquiring two-dimensional slices of a three-dimensional input image frame by frame, takes a two-dimensional subgraph as input of a two-dimensional segmentation network, and the two-dimensional segmentation network obtains finer local differences from slice layers;
A4、and Q is equal to out Obtaining->And Q is equal to in Obtained through a fusion moduleThe fusion module is used for fusing the three-dimensional model output and the two-dimensional model output, and a weight regulator, Q is applied out ，/>And->The initial weights of (a) are α=0.4, β=0.3, γ=0.3 and then the three weight parameters are automatically adjusted by the back propagation algorithm, by calculating the loss functionAnd (3) carrying out parameter updating by using gradient of the number relative to the weight parameter and utilizing optimization algorithms such as gradient descent and the like.
A5、Respectively with Y out And Y in Comparing and calculating loss to obtain L out And L in Two segmentation loss functions.
The specific process of the image block contrast learning task is as follows:
b1, for two marker imagesAnd->And two unlabeled images->And->X obtained by bidirectional copy and paste out And X in Performing blocking treatment;
b2, inputting the obtained small blocks into a 3D U-Net teacher model and a 3D U-Net student model respectively to obtain output;
b3, calculating the block contrast learning loss functionAnd->The outputs of the same block through the 3D U-Net teacher model and the 3D U-Net student model are directly opposite, and vice versa.
Step 2, summing the loss functions obtained in the step 1 to obtain a total loss function; the specific calculation expression is as follows:
wherein L is all Represents the total loss function, L in And L out The segmentation loss between the representation and the label,and->Representing block contrast learning penalty, M is a mask, < ->And->Representing a fusion result of the three-dimensional segmentation result and the two-dimensional segmentation result, < >>Y out And Y in Representing a mixed label obtained after the two-way copy pasting of a pseudo label and a real label, L seg Is a linear combination of the Dice loss and the cross entropy loss, and the Dice loss calculation formula is +.>Wherein |X| and |Y| respectively represent the real label and the prediction mask of the segmentation, N represents the number of samples, m represents the number of image blocks, d is the embedding distance, and the calculation formula is X represents out Results of each pixel of (a) output by the student model, a->X represents out Results of the respective pixels output by the teacher model, and (2)>X represents in Results of each pixel of (a) output by the student model, a->X represents in Results of each pixel of (3) output by the teacher model.
Step 3, updating a student model of the 3D medical image segmentation network by using random gradient descent of the total loss function;
and 4, repeating the steps 1 to 3 until the total loss function is not reduced, and optimizing the 3D medical image segmentation.
Therefore, the 3D medical image segmentation method based on the multi-dimensional and global local combination is adopted, a 3D teacher model and a 2D teacher model are trained through labeled data, bidirectional copy and paste of unlabeled data and labeled data are used as input of a 3D student model, image slices after bidirectional copy and paste are used as input of a 2D student model, unlabeled images are used as input of the 3D teacher model to obtain pseudo labels, and the pseudo labels and the real labels are subjected to bidirectional copy and paste to obtain mixed labels, the mixed labels are used for supervising the mixed output of the 3D student model and the 2D student model, meanwhile, auxiliary task image block contrast learning is added to improve performance, and a student model capable of accurately segmenting the 3D medical image is trained through random gradient reduction of total loss.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention and not for limiting it, and although the present invention has been described in detail with reference to the preferred embodiments, it will be understood by those skilled in the art that: the technical scheme of the invention can be modified or replaced by the same, and the modified technical scheme cannot deviate from the spirit and scope of the technical scheme of the invention.
Claims (7)
1. A 3D medical image segmentation method based on multi-dimensional and global local association, comprising the steps of:
step 1, processing an unlabeled image and a marked image through two parallel tasks to obtain a corresponding loss function;
step 2, summing the loss functions obtained in the step 1 to obtain a total loss function;
step 3, updating a student model of the 3D medical image segmentation network by using random gradient descent of the total loss function;
and 4, repeating the steps 1 to 3 until the total loss function is not reduced, and optimizing the 3D medical image segmentation.
2. A 3D medical image segmentation method based on a combination of multi-dimensions and global localization as claimed in claim 1, characterized in that: in the step 1, two parallel tasks comprise a multidimensional joint segmentation task and an image block comparison learning task.
3. The 3D medical image segmentation method based on multi-dimensional and global local joint according to claim 2, wherein the specific process of the multi-dimensional joint segmentation task is as follows:
a1, two unlabeled imagesAnd->Inputting a 3D U-Net teacher model and obtaining a reliable pseudo tag by a multi-level confidence coefficient module>And->The pseudo tag and the real tag are copied and stuck in two directions to obtain a mixed tag Y out And Y in ；
A2, two marker imagesAnd->And two unlabeled images->And->X is obtained through bidirectional copy and paste out And X in Input 3D U-Net student model, output Q out And Q in ；
A3, X is out And X in Input two-dimensional consistency learning module to obtainAnd->
A4、And Q is equal to out Obtaining-> And Q is equal to in Obtaining->
A5、Respectively with Y out And Y in Comparing and calculating loss to obtain L out And L in Two segmentation loss functions.
4. A 3D medical image segmentation method based on multi-dimensional and global local association according to claim 3, wherein the specific process of the image block contrast learning task is as follows:
b1, for two marker imagesAnd->And two unlabeled images->And->X obtained by bidirectional copy and paste out And X in Performing blocking treatment;
b2, inputting the obtained small blocks into a 3D U-Net teacher model and a 3D U-Net student model respectively to obtain output;
b3, calculating the block contrast learning loss functionAnd->The outputs of the same block through the 3D U-Net teacher model and the 3D U-Net student model are directly opposite, and vice versa.
5. The method for 3D medical image segmentation based on multi-dimensional and global local integration according to claim 4, wherein the calculation expression of the total loss function in step 2 is as follows:
wherein L is all Represents the total loss function, L in And L out The segmentation loss between the representation and the label,and->Representing block contrast learning penalty, M is a mask, < ->And->Representing a fusion result of the three-dimensional segmentation result and the two-dimensional segmentation result,Y out and Y in Representing a mixed label obtained after the two-way copy pasting of a pseudo label and a real label, L seg Is a linear combination of the Dice loss and the cross entropy loss, and the Dice loss calculation formula is +.>Wherein |X| and |Y| respectively represent the real label and the prediction mask of the segmentation, N represents the number of samples, m represents the number of image blocks, d is the embedding distance, and the calculation formula is X represents out Results of each pixel of (a) output by the student model, a->X represents out Results of the respective pixels output by the teacher model, and (2)>X represents in Results of each pixel of (a) output by the student model, a->X represents in Results of each pixel of (3) output by the teacher model.
6. The method for 3D medical image segmentation based on multi-dimensional and global local integration according to claim 5, wherein the two unlabeled images in A1And->Inputting a 3D U-Net teacher model and obtaining a reliable pseudo tag by a multi-level confidence coefficient module>And->The specific process of (2) is as follows: the 3D U-Net teacher model comprises an encoder and a decoder, wherein the decoder comprises a plurality of layers of networks, the output of each layer of the three layers is regulated to be the same by using bilinear interpolation, and the channel pooling is used for regulating the channel number of each layer of output to be equal; and obtaining a predicted value by utilizing the Monte Carlo Dropout, wherein the entropy has a fixed numerical range, measuring the multi-scale uncertainty u by utilizing the predicted entropy, and obtaining a reliable pseudo tag by utilizing u as an adaptive weight and correspondingly if the accuracy of pseudo tag prediction is high, wherein the occupied weight is large.
7. The multi-dimensional and global local joint-based 3D medical image segmentation method according to claim 6, wherein: in A3, the two-dimensional consistency learning module performs dimension reduction operation by acquiring two-dimensional slices of the three-dimensional input image frame by frame, takes the two-dimensional subgraph as input of a two-dimensional segmentation network, and the two-dimensional segmentation network obtains finer local differences from slice layers.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311852915.0A CN117830324B (en) | 2023-12-28 | 2023-12-28 | 3D medical image segmentation method based on multi-dimensional and global local combination |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311852915.0A CN117830324B (en) | 2023-12-28 | 2023-12-28 | 3D medical image segmentation method based on multi-dimensional and global local combination |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117830324A true CN117830324A (en) | 2024-04-05 |
| CN117830324B CN117830324B (en) | 2024-07-12 |
Family
ID=90520362
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311852915.0A Active CN117830324B (en) | 2023-12-28 | 2023-12-28 | 3D medical image segmentation method based on multi-dimensional and global local combination |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117830324B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110826458A (en) * | 2019-10-31 | 2020-02-21 | 河海大学 | Multispectral remote sensing image change detection method and system based on deep learning |
| CN112801107A (en) * | 2021-02-01 | 2021-05-14 | 联想(北京)有限公司 | Image segmentation method and electronic equipment |
| US20220309674A1 (en) * | 2021-03-26 | 2022-09-29 | Nanjing University Of Posts And Telecommunications | Medical image segmentation method based on u-net |
| CN115331009A (en) * | 2022-08-17 | 2022-11-11 | 西安理工大学 | Medical image segmentation method based on multitask MeanTeacher |
| CN115861164A (en) * | 2022-09-16 | 2023-03-28 | 重庆邮电大学 | Medical image segmentation method based on multi-field semi-supervision |
| CN116468746A (en) * | 2023-03-27 | 2023-07-21 | 华东师范大学 | Bidirectional copy-paste semi-supervised medical image segmentation method |
| CN116664588A (en) * | 2023-05-29 | 2023-08-29 | 华中科技大学 | Mask modeling-based 3D medical image segmentation model building method and application thereof |
-
2023
- 2023-12-28 CN CN202311852915.0A patent/CN117830324B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110826458A (en) * | 2019-10-31 | 2020-02-21 | 河海大学 | Multispectral remote sensing image change detection method and system based on deep learning |
| CN112801107A (en) * | 2021-02-01 | 2021-05-14 | 联想(北京)有限公司 | Image segmentation method and electronic equipment |
| US20220309674A1 (en) * | 2021-03-26 | 2022-09-29 | Nanjing University Of Posts And Telecommunications | Medical image segmentation method based on u-net |
| CN115331009A (en) * | 2022-08-17 | 2022-11-11 | 西安理工大学 | Medical image segmentation method based on multitask MeanTeacher |
| CN115861164A (en) * | 2022-09-16 | 2023-03-28 | 重庆邮电大学 | Medical image segmentation method based on multi-field semi-supervision |
| CN116468746A (en) * | 2023-03-27 | 2023-07-21 | 华东师范大学 | Bidirectional copy-paste semi-supervised medical image segmentation method |
| CN116664588A (en) * | 2023-05-29 | 2023-08-29 | 华中科技大学 | Mask modeling-based 3D medical image segmentation model building method and application thereof |
Non-Patent Citations (1)
| Title |
|---|
| 刘静等: "基于改进的FLICM的医学图像分割研究", 微型机与应用, vol. 35, no. 23, 31 December 2016 (2016-12-31) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117830324B (en) | 2024-07-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108428229B (en) | Lung texture recognition method based on appearance and geometric features extracted by deep neural network | |
| WO2022001623A1 (en) | Image processing method and apparatus based on artificial intelligence, and device and storage medium | |
| WO2018023734A1 (en) | Significance testing method for 3d image | |
| CN109903301B (en) | Image contour detection method based on multistage characteristic channel optimization coding | |
| CN107247971B (en) | Intelligent analysis method and system for ultrasonic thyroid nodule risk index | |
| CN110276402B (en) | Salt body identification method based on deep learning semantic boundary enhancement | |
| CN111445474B (en) | Kidney CT image segmentation method based on bidirectional re-attention depth network | |
| CN102542302A (en) | Automatic complicated target identification method based on hierarchical object semantic graph | |
| CN115761222B (en) | Image segmentation method, remote sensing image segmentation method and device | |
| CN113111716B (en) | Remote sensing image semiautomatic labeling method and device based on deep learning | |
| CN104517120B (en) | Orthogonal matched remote sensing images scene classification method is layered based on multichannel | |
| CN110866921A (en) | Weakly supervised vertebral body segmentation method and system based on self-training and slice propagation | |
| CN117557414B (en) | Cultivated land supervision method, device, equipment and storage medium based on automatic interpretation of remote sensing image | |
| CN116486408B (en) | Cross-domain semantic segmentation method and device for remote sensing image | |
| CN114283285A (en) | Cross consistency self-training remote sensing image semantic segmentation network training method and device | |
| CN115966302A (en) | Semi-supervised gonitis auxiliary analysis method based on deep contrast learning | |
| CN117523194A (en) | Image segmentation method based on sparse labeling | |
| CN113591633A (en) | Object-oriented land utilization information interpretation method based on dynamic self-attention Transformer | |
| Li et al. | Automatic annotation algorithm of medical radiological images using convolutional neural network | |
| CN110533074B (en) | Automatic image category labeling method and system based on double-depth neural network | |
| CN115760869A (en) | Attention-guided non-linear disturbance consistency semi-supervised medical image segmentation method | |
| CN117152427A (en) | Remote sensing image semantic segmentation method and system based on diffusion model and knowledge distillation | |
| Nakhaee et al. | DeepRadiation: An intelligent augmented reality platform for predicting urban energy performance just through 360 panoramic streetscape images utilizing various deep learning models | |
| CN102609721B (en) | Remote sensing image clustering method | |
| CN113052759B (en) | Scene complex text image editing method based on MASK and automatic encoder |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |