CN116934759A - Local correction interactive medical image segmentation method and system - Google Patents
Local correction interactive medical image segmentation method and system Download PDFInfo
- Publication number
- CN116934759A CN116934759A CN202311200851.6A CN202311200851A CN116934759A CN 116934759 A CN116934759 A CN 116934759A CN 202311200851 A CN202311200851 A CN 202311200851A CN 116934759 A CN116934759 A CN 116934759A
- Authority
- CN
- China
- Prior art keywords
- segmentation
- feedback
- feature map
- local
- global
- 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
Links
- 238000012937 correction Methods 0.000 title claims abstract description 48
- 230000002452 interceptive effect Effects 0.000 title claims abstract description 37
- 238000003709 image segmentation Methods 0.000 title claims abstract description 28
- 238000000034 method Methods 0.000 title claims abstract description 21
- 230000011218 segmentation Effects 0.000 claims abstract description 161
- 230000003993 interaction Effects 0.000 claims abstract description 34
- 238000013507 mapping Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 9
- 238000005192 partition Methods 0.000 claims description 7
- 238000007670 refining Methods 0.000 claims description 7
- 238000010586 diagram Methods 0.000 claims description 2
- 238000012549 training Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 230000004927 fusion Effects 0.000 description 2
- 238000010191 image analysis Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 206010028980 Neoplasm Diseases 0.000 description 1
- 210000003484 anatomy Anatomy 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000003902 lesion Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000001915 proofreading effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 238000011282 treatment 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
- G06T7/0002—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
-
- 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]
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a local correction interactive medical image segmentation method and a system in the technical field of medical image segmentation, aiming at solving the problems of low precision, large operand, high cost and the like of the medical image automatic segmentation method in the prior art when aiming at the difficult segmentation task of few samples, comprising the following steps of carrying out the first step with a user after obtaining a primary automatic segmentation resulttPerforming secondary interaction; if the user judges that the segmentation result is inaccurate, performing correction interactive clicking, generating a patch according to clicking, cutting the corresponding areas of the patch on the trunk feature map and the previous segmentation mask, and performing local refinement and global correction to obtain the first steptAnd submitting the segmentation result after secondary interaction refinement correction to user judgment again, repeating the steps until the segmentation result is accurate, and outputting the segmentation result. The invention adds a local correction module based on interaction on the basis of automatic segmentation to refine the local and correct the global of the automatic segmentation resultAnd (5) improving the segmentation accuracy.
Description
Technical Field
The invention relates to a local correction interactive medical image segmentation method and system, and belongs to the technical field of medical image segmentation.
Background
Medical image segmentation refers to the automatic or semi-automatic separation of anatomical structures or lesion areas of interest to a user from a medical image. It is an important step in medical image analysis, one of the fundamental technologies for various medical applications, and has important significance for medical diagnosis, treatment and research.
The main pursuit of the medical image segmentation method is to improve segmentation accuracy and further reduce medical image labeling cost. With the development of deep learning, highly supervised approaches have been able to achieve performance levels approaching that of human experts, but a large number of disease-specific segmentation tasks have not achieved enough segmentation labels for model training.
In order to solve the problems that an automatic segmentation model is difficult to realize in a medical image segmentation task and the obtained result is poor, a learner tries to optimize the segmentation result by adopting a method of introducing an interactive segmentation correction module.
Prior document 1 (Luo X, wang G, song T, zhang J, et al, mideeppseg: minimally interactive segmentation of unseen objects from medical images using deep learning [ J ]. Medical Image Analysis, 2021, 72: 102102) proposes a new method for interactive segmentation of medical images based on deep learning, which encodes internal edge points provided by a user by exponential ranging, enabling CNN to achieve good initial segmentation results for previously seen and unseen objects, and in particular has good versatility for a series of previously unseen objects. However, the advantages of the automatic segmentation model are almost completely abandoned, the model is completely guided by correction depending on user judgment and interaction operation, and the accurate segmentation of the model needs to depend on a large amount of interaction operation.
In the prior art, in document 2 (Liu W, ma C, yang Y, et al Transforming the Interactive Segmentation for Medical Imaging [ C ]. International Conference on Medical Image Computing and Computer Assisted Intervention, 2022), an interaction module is added after nnU-Net through a transducer, and the automatic segmentation result is corrected through user interaction, so that good performance is obtained on part of the data set. Although an interaction module is added, the user interaction is simply taken as a sample to carry out full-image proofreading, and the specificity of medical images and such segmentation tasks is not considered. However, unlike the conventional natural image, when dealing with the medical image segmentation task of a disease region such as a tumor, the focal region to be segmented is usually small and concentrated, so that a large amount of calculation force is wasted on the calibration of the non-focal position due to the simple whole image iteration, and the segmentation result is also easily interfered by the non-disease region.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, and provides a local correction interactive medical image segmentation method and a local correction interactive medical image segmentation system.
In order to achieve the above purpose, the invention is realized by adopting the following technical scheme:
in one aspect, the present invention provides a local correction interactive medical image segmentation method comprising the steps of:
s1, acquiring a target image, importing the target image into an encoder network, mapping the target image to a segmentation mask, and acquiring a primary segmentation result;
s2, carrying out the first step on the segmentation result with the usertPerforming secondary interaction to obtain judgment of a user on a segmentation result;
s3, if the judgment of the segmentation result by the user is inaccurate, the user performs annotation click on the image, patches are generated by taking the user annotation click as the center, the trunk feature map and the patch area on the previous segmentation mask are cut, the patch area is locally thinned, and local segmentation feedback after local thinning is obtained;
s4, pasting the locally refined local segmentation feedback back to the previous segmentation mask to obtain locally refined global segmentation feedback; after the global segmentation feedback is updated according to the locally refined global segmentation feedback,fusing the main feature map with the main feature map to obtain the firsttRefining the corrected segmentation result by secondary interaction;
s5, carrying out the first step with the usertAnd (5) carrying out +1 interaction, and repeating the steps S3 and S4 until the judgment of the segmentation result by the user is accurate, and outputting the segmentation result.
Further, the target image is imported into the encoder network, mapped to the segmentation mask, and expressed as a primary segmentation result as:
；
wherein ,automatic segmentation mask for encoder network feedback, < >>Embedding for dense features->For encoder module->Is the target image, and->，RRepresents a set of real numbers,H、W、Drespectively representing the height, width, number of channels, < >>Is a super parameter.
Further, whenAt the time, the previous division mask is +.>。
Further, the step S3 specifically includes:
recording the position coordinates of the user annotation click asThe type tag record of comment click is +.>；
Converting user annotation clicks into interactive mappings by disk encoding module，/>, wherein tRepresent the firsttAn interaction map obtained from the secondary interactions,Rrepresents a set of real numbers,HandWrespectively representing the height and width of the image, 2 representing the number of channels;
cascading the target image and the interaction map on the depth level to obtain a trunk feature map, wherein /> and />The height and the width of the feature map are respectively represented, and 3 represents the number of channels;
generating a rectangular patch by taking user annotation click as a center, and cutting out the corresponding position from the trunk feature map to obtain a trunk feature map patch blockThe size is +.>, wherein rIs the expansion ratio; at the same time, the corresponding position of the rectangular patch is masked from the previous division>Cutting out the patch block to obtain the previous divided patch block->；
Patch block for trunk feature mapF C Performing pixel-by-pixel matching, measuring backbone feature map patchF C Feature affinity between each pixel point in (a) and user annotation clickThe expression is:
；
wherein ,for pixel coordinates, +.>Coordinates of clicks are annotated for the user in the backbone feature map patch,performing regular operation for L2;
according to the characteristic affinityFor the previous divided patch +.>And Label of comment click->Linear combination is carried out to generate local segmentation feedback after local refinement, and the expression is as follows:
；
wherein ,for local segmentation feedback->For Hadamard product operation, +.>Is a similarity threshold.
Further, after updating the global segmentation feedback after local refinement, fusing the global segmentation feedback with a trunk feature map to obtain the firsttRefining the corrected segmentation result by secondary interaction, comprising:
using convolution blocksFor trunk feature mapFCoding, then cascading the coded backbone feature map with the locally refined global segmentation feedback channel in the channel dimension, and utilizing a convolution block +.>And updating global segmentation feedback by using a Sigmoid function, wherein the expression is as follows:
；
wherein ,for updated global partition feedback, +.>、/>For convolution block +.>、/>Convolution blocks respectively->、/>Is a learning parameter of->Feedback for global segmentation->Is a cascade of channels;
by convolving blocksFusing the trunk feature map with updated segmentation feedback to obtain a fused trunk feature map, wherein the expression is as follows:
；
wherein ,to fuse the backbone feature map, < >>Is->Is a learning parameter of (a);
will fuse the backbone feature mapAnd updated segmentation feedback +.>Respectively substitute for backbone feature mapFAnd global segmentation feedback->Obtaining a new segmentation mask which is output after the interactive correction>Namely the firsttAnd refining the corrected segmentation result by secondary interaction.
Further, if the segmentation result is accurate, outputting the segmentation result in the form of a segmentation mask and a superimposed image, storing the corresponding segmentation mask obtained by the completed segmentation algorithm as a new sample set, and continuously training the automatic segmentation correction module by a fine adjustment incremental learning module on the new sample set after the specified number is reached.
In another aspect, the present invention provides a locally-rectified interactive medical image segmentation system for implementing a locally-rectified interactive medical image segmentation method as defined in any one of the above.
Optionally, the segmentation system includes an encoder module, a disk encoding module and a segmentation rectification module, the encoder module is used for mapping the imported target image to a segmentation mask, and the disk encoding module is used for converting the user annotation click into an interactive mapping;
the segmentation correction module is used for interactive correction and comprises a local correction module, and the local correction module is used for carrying out local refinement and global correction on the target image.
Optionally, the segmentation correction module is required to perform pre-training, where the pre-training uses a loss functionLThe expression is:
；
wherein ,to normalize the focus loss function +.>Feedback for global segmentation->For a true segmentation mask,for updated global partition feedback, +.>A mask for the new partition.
Optionally, the encoder module employs nnU-Net and the segmentation correction module employs a full convolution baseline architecture.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a local correction module which performs local refinement on a segmentation result based on user interaction and then performs global correction, so that the segmentation precision is improved in the process, and the operation amount and interaction times can be effectively reduced and the segmentation cost is reduced relative to global refinement correction;
the invention is integrated with the increment learning module, so that the model can update and adjust the encoder module in real time through the actual segmentation task, and the segmentation accuracy of primary segmentation is further improved.
Drawings
Fig. 1 is a flow chart of a method for locally correcting an interactive medical image segmentation in an embodiment of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples are only for more clearly illustrating the technical aspects of the present invention, and are not intended to limit the scope of the present invention.
Example 1
As shown in fig. 1, the method for segmenting a local correction interactive medical image according to the embodiment of the invention includes the following steps:
s1, importing a target image into an encoder module, wherein the encoder module adopts nnU-Net, the imported target image is mapped to a segmentation mask, and is used as a primary segmentation result, the primary segmentation result is an automatic segmentation result, and the expression is as follows:
；
wherein ,y mask for an automatic segmentation mask for encoder network feedback,embedding for dense features->In order for the encoder module to be a function of the encoder module,Iis the target image, and->，RRepresents a set of real numbers,H、W、Drespectively representing the height, width, number of channels, < >>Is a super parameter.
S2, carrying out the first step with the usertThe next interaction, the previous segmentation mask is covered by the userAnd judging whether the segmentation result is accurate or not. It should be noted that for the first interaction, since the initial segmentation is derived from the automatic segmentation, when +.>At the time of previous division mask->Is->。
If the segmentation result is accurate, ending the segmentation calculation, outputting the segmentation result in the form of a segmentation mask and a superimposed image, storing the corresponding segmentation mask obtained by the completed segmentation calculation as a new sample set, and continuously training the encoder module by a fine adjustment increment learning module on the new sample set after the number of the new sample set is up to a specified number, so that the encoder module can be updated and adjusted in real time through an actual segmentation task, and the segmentation accuracy of primary segmentation is further improved.
S3, if the segmentation result is inaccurate, interactive correction is needed, and a user inputs a target imageAnd the previous segmentation mask->Annotating the image, recording the position coordinates of the annotating click as +.>The type tag record of comment click is +.>。
Conversion of annotation clicks into interactive mappings by disk encoding module, wherein tRepresent the firsttInteraction map from secondary interactions +.>，RRepresents a set of real numbers,HandWthe height and width of the image are shown, respectively, and 2 represents the number of channels.
S4, inputting a target imageAnd (d)tInteraction map of secondary interactions->Delivering into trunk, and cascading at depth level to obtain trunk feature map +.>, wherein /> and />The height and width of the feature map are indicated, respectively, and 3 indicates the number of channels.
Previous segmentation maskThen the obtained product is transmitted to a local correction module, firstly, the local refinement is carried out, the local correction module generates a rectangular patch by taking the click of user comments as the center, and the corresponding position of the rectangular patch is obtained from the following positionFeature map generated by trunkFCutting out to obtain a trunk feature map patch block +.>The size is +.>, wherein rIs the expansion ratio. At the same time, the corresponding position of the rectangular patch is masked from the previous division>Cutting out the patch block to obtain the previous divided patch block->。
Patch block for trunk feature mapF C Performing pixel-by-pixel matching, measuring backbone feature map patchF C Feature affinity between each pixel point in (a) and user annotation clickThe feature affinity is defined as cosine similarity, and the expression is:
；
wherein ,the coordinates of the pixel points; but->Is the coordinates of the user annotation click in the backbone feature map patch, +.>Performing regular operation for L2; the greater the feature affinity, the more likely a pixel is to belong to the same category as the annotation click.
According to the characteristic affinityFor a pair ofThe patch block is split up before->And Label of comment click->Performing linear combination to generate partial segmentation feedback after partial refinement>The expression is:
；
wherein ,for Hadamard product operation, +.>Is a similarity threshold.
S5, performing global correction and feeding back the local segmentationPaste back the previous split mask->Cutting position on the upper part to obtain global segmentation feedback after local refinement>。
Using convolution blocksFor backbone feature map->Coding, then coding the main trunk characteristic diagram +.>Global segmentation feedback after local refinement in channel dimension>Channel concatenation and use of convolution block +.>And Sigmoid function feedback on global segmentation +.>Updating, wherein the expression is as follows:
；
wherein ,for updated global partition feedback, +.>、/>For convolution block +.>、/>Respectively->、/>Is a learning parameter of->Is a cascade of channels.
Using updated segmentation feedbackFor trunk feature mapFUpdating: by convolution block->Mapping trunk characteristicsFAnd updated segmentation feedback->Fusion to obtain a fusion backbone profile->The expression is:
；
wherein ,is->Is provided.
Will fuse the backbone feature mapAnd updated segmentation feedback +.>Respectively substitute for backbone feature mapFAnd global segmentation feedback->Obtaining a new segmentation mask which is output after the interactive correction>As the firsttAnd (5) a segmentation result of the secondary interaction until the segmentation result is accurate.
In addition, a loss function is required to be adopted for the segmentation correction moduleLPretraining is performed, and the expression is as follows:
；
wherein ,to normalize the focus loss function +.>Is the true segmentation mask.
Example 2
The embodiment of the invention provides a local correction interactive medical image segmentation system, which comprises an encoder module, a disk encoding module and a segmentation correction module, wherein the encoder module adopts nnU-Net for mapping an imported target image to a segmentation mask, and the disk encoding module is used for converting user annotation click into interaction mapping.
The segmentation correction module adopts a full convolution baseline architecture and is used for carrying out interactive correction on an imported target image, and the full convolution baseline architecture comprises a trunk and a segmentation head, wherein the trunk is used for capturing long-range dependence of pixels to generate advanced features, the segmentation head is used for recovering spatial resolution and extracting semantic information, a local correction module is inserted between the trunk and the segmentation head, and the local correction module is used for carrying out local refinement and global correction on the target image.
The foregoing is merely a preferred embodiment of the present invention, and it should be noted that modifications and variations could be made by those skilled in the art without departing from the technical principles of the present invention, and such modifications and variations should also be regarded as being within the scope of the invention.
Claims (10)
1. A method of locally-rectified interactive medical image segmentation, comprising:
s1, acquiring a target image, importing the target image into an encoder network, mapping the target image to a segmentation mask, and acquiring a primary segmentation result;
s2, carrying out the first step on the segmentation result with the usertPerforming secondary interaction to obtain judgment of a user on a segmentation result;
s3, if the judgment of the segmentation result by the user is inaccurate, the user performs annotation click on the image, patches are generated by taking the user annotation click as the center, the trunk feature map and the patch area on the previous segmentation mask are cut, the patch area is locally thinned, and local segmentation feedback after local thinning is obtained;
s4, pasting the locally refined local segmentation feedback back to the previous segmentation mask to obtain locally refined global segmentation feedback; updating the global segmentation feedback according to the global segmentation feedback after the local refinement, and fusing the global segmentation feedback with the trunk feature map to obtain the firsttRefining the corrected segmentation result by secondary interaction;
s5, carrying out the first step with the usertAnd (5) carrying out +1 interaction, and repeating the steps S3 and S4 until the judgment of the segmentation result by the user is accurate, and outputting the segmentation result.
2. The locally corrected interactive medical image segmentation method according to claim 1, wherein the target image is imported into an encoder network, mapped to a segmentation mask, and as a primary segmentation result, expressed as:
；
wherein ,y mask for an automatic segmentation mask for encoder network feedback,embedding for dense features->In order for the encoder module to be a function of the encoder module,Iis the target image, and->，RRepresents a set of real numbers,H、W、Drespectively representing the height, width and channel number of the image,is a super parameter.
3. The locally-rectified interactive medical image segmentation method according to claim 1, which is specificCharacterized in that whenAt the time, the previous division mask is +.>，/>An automatic segmentation mask for encoder network feedback.
4. The locally corrected interactive medical image segmentation method according to claim 1, wherein step S3 specifically comprises:
recording the position coordinates of the user annotation click asThe type tag record of comment click is +.>；
Converting user annotation clicks into interaction mappings，/>, wherein tRepresent the firsttAn interaction map obtained from the secondary interactions,Rrepresents a set of real numbers,HandWrespectively representing the height and width of the image, 2 representing the number of channels;
cascading the target image and the interaction map on the depth level to obtain a trunk feature map, wherein /> and />The height and the width of the feature map are respectively represented, and 3 represents the number of channels;
generating a rectangular patch by taking user annotation click as a center, and cutting out the corresponding position from the trunk feature map to obtain a trunk feature map patch blockThe size is +.>, wherein rIs the expansion ratio; at the same time, the corresponding position of the rectangular patch is masked from the previous division>Cutting out the patch block to obtain the previous divided patch block->；
Patch block for trunk feature mapF C Performing pixel-by-pixel matching, measuring backbone feature map patchF C Feature affinity between each pixel point in (a) and user annotation clickThe expression is:
；
wherein ,for pixel coordinates, +.>Annotating the coordinates of the click for the user in the backbone feature map patch, +.>Performing regular operation for L2;
according to the characteristic affinityFor the previous divided patch +.>And Label of comment click->Linear combination is carried out to generate local segmentation feedback after local refinement, and the expression is as follows:
；
wherein ,feedback for local segmentation after local refinement, +.>For Hadamard product operation, +.>Is a similarity threshold.
5. The local correction interactive medical image segmentation method according to claim 1, wherein after updating the locally refined global segmentation feedback, fusing the locally refined global segmentation feedback with a trunk feature map to obtain a first feature maptRefining the corrected segmentation result by secondary interaction, comprising:
using convolution blocksFor trunk feature mapFCoding, cascading the coded backbone feature diagram with the locally refined global segmentation feedback channel in the channel dimension, and utilizingConvolution block->And updating global segmentation feedback by using a Sigmoid function, wherein the expression is as follows:
；
wherein ,for updated global partition feedback, +.>、/>For convolution block +.>、/>Convolution blocks respectively->、/>Is a learning parameter of->Feedback for global segmentation->Is a cascade of channels;
by convolving blocksFusing the trunk feature map with the updated segmentation feedback to obtain a fused trunk feature map,the expression is:
；
wherein ,to fuse the backbone feature map, < >>Is->Is a learning parameter of (a);
will fuse the backbone feature mapAnd updated global partition feedback +.>Respectively substitute for backbone feature mapFAnd global segmentation feedback->Obtaining a new segmentation mask which is output after the interactive correction>Namely the firsttRefining the corrected segmentation result by secondary interaction, wherein,Rrepresents a set of real numbers,HandWrepresenting the height and width of the image, respectively.
6. The local correction interactive medical image segmentation method according to claim 1, wherein if the user judges that the segmentation result is accurate, the segmentation result is output in the form of a segmentation mask and a superimposed image, and the corresponding segmentation mask obtained by the completed segmentation algorithm is stored as a new sample set, and after the specified number is reached, the automatic segmentation correction module is continuously trained by the fine adjustment incremental learning module on the new sample set.
7. A locally-rectified interactive medical image segmentation system, characterized in that the segmentation system is configured to implement the locally-rectified interactive medical image segmentation method according to any one of claims 1-6.
8. The locally rectified interactive medical image segmentation system according to claim 7, wherein the segmentation system comprises an encoder module for mapping the imported target image to a segmentation mask, a disk encoding module for converting user annotation clicks into an interaction map, and a segmentation rectification module;
the segmentation correction module is used for interactively correcting the target image and comprises a local correction module, and the local correction module is used for locally refining and globally correcting the target image.
9. The locally corrected interactive medical image segmentation system of claim 8, wherein the segmentation correction module is pre-trained using a loss functionLThe expression is:
；
wherein ,to normalize the focus loss function +.>Feedback for global segmentation->For true segmentation mask, < >>To updatePost global segmentation feedback,/>A mask for the new partition.
10. The locally corrected interactive medical image segmentation system of claim 8, wherein the encoder module employs nnU-Net and the segmentation correction module employs a full convolution baseline architecture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311200851.6A CN116934759A (en) | 2023-09-18 | 2023-09-18 | Local correction interactive medical image segmentation method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311200851.6A CN116934759A (en) | 2023-09-18 | 2023-09-18 | Local correction interactive medical image segmentation method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116934759A true CN116934759A (en) | 2023-10-24 |
Family
ID=88381104
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311200851.6A Pending CN116934759A (en) | 2023-09-18 | 2023-09-18 | Local correction interactive medical image segmentation method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116934759A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170178349A1 (en) * | 2015-12-18 | 2017-06-22 | The Johns Hopkins University | Method for deformable 3d-2d registration using multiple locally rigid registrations |
| CN114037712A (en) * | 2021-11-01 | 2022-02-11 | 广州凡科互联网科技股份有限公司 | User real-time smearing interactive image segmentation method based on deep learning |
| CN116433914A (en) * | 2023-04-28 | 2023-07-14 | 齐鲁工业大学(山东省科学院) | Two-dimensional medical image segmentation method and system |
-
2023
- 2023-09-18 CN CN202311200851.6A patent/CN116934759A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170178349A1 (en) * | 2015-12-18 | 2017-06-22 | The Johns Hopkins University | Method for deformable 3d-2d registration using multiple locally rigid registrations |
| CN114037712A (en) * | 2021-11-01 | 2022-02-11 | 广州凡科互联网科技股份有限公司 | User real-time smearing interactive image segmentation method based on deep learning |
| CN116433914A (en) * | 2023-04-28 | 2023-07-14 | 齐鲁工业大学(山东省科学院) | Two-dimensional medical image segmentation method and system |
Non-Patent Citations (3)
| Title |
|---|
| QIAOQIAO WEI 等: "Focused and Collaborative Feedback Integration for Interactive Image Segmentation", 《CVF》, pages 18643 - 18652 * |
| WENTAO LIU 等: "Transforming the Interactive Segmentation for Medical Imaging", 《MEDICAL IMAGE COMPUTING AND COMPUTER ASSISTED INTERVENTION》, pages 704 - 713 * |
| 贲雅芳: "基于差异性点击策略的交互式图像分割算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》, no. 01, pages 138 - 2199 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113012172B (en) | AS-UNet-based medical image segmentation method and system | |
| Zhang et al. | A comprehensive survey on segment anything model for vision and beyond | |
| CN112102237A (en) | Brain tumor recognition model training method and device based on semi-supervised learning | |
| CN113240691A (en) | Medical image segmentation method based on U-shaped network | |
| Cho et al. | A large RGB-D dataset for semi-supervised monocular depth estimation | |
| CN111325750B (en) | Medical image segmentation method based on multi-scale fusion U-shaped chain neural network | |
| Xu et al. | Accurate and efficient stereo matching via attention concatenation volume | |
| CN116434033A (en) | Cross-modal contrast learning method and system for RGB-D image dense prediction task | |
| Zhou et al. | PADENet: An efficient and robust panoramic monocular depth estimation network for outdoor scenes | |
| Tian et al. | RGB oralscan video-based orthodontic treatment monitoring | |
| CN113538363A (en) | Lung medical image segmentation method and device based on improved U-Net | |
| CN114240811A (en) | Method for generating new image based on multiple images | |
| Bauer et al. | Weakly supervised multi-modal 3d human body pose estimation for autonomous driving | |
| CN116934759A (en) | Local correction interactive medical image segmentation method and system | |
| Yang et al. | DRE-Net: A dynamic radius-encoding neural network with an incremental training strategy for interactive segmentation of remote sensing images | |
| CN117237322A (en) | Organ segmentation modeling method and terminal based on medical image | |
| CN117392289A (en) | Method and system for automatically generating case field video based on AI (advanced technology attachment) voice | |
| Zhang et al. | VPUFormer: Visual prompt unified transformer for interactive image segmentation | |
| CN111582449A (en) | Training method, device, equipment and storage medium for target domain detection network | |
| CN116523926A (en) | Tissue precise segmentation modeling method and device for multi-task bone joint CT/MRI fusion | |
| CN116310335A (en) | Method for segmenting pterygium focus area based on Vision Transformer | |
| Song et al. | Bi-directional seed attention network for interactive image segmentation | |
| Qian et al. | Context-aware transformer for 3d point cloud automatic annotation | |
| Huang et al. | Deep learning methods in image matting: a survey | |
| Wang et al. | Contact-conditioned hand-held object reconstruction from single-view images |
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 |