[go: up one dir, main page]

CN117132776A - A multi-organ image segmentation model construction method and segmentation method - Google Patents

A multi-organ image segmentation model construction method and segmentation method Download PDF

Info

Publication number
CN117132776A
CN117132776A CN202311174265.9A CN202311174265A CN117132776A CN 117132776 A CN117132776 A CN 117132776A CN 202311174265 A CN202311174265 A CN 202311174265A CN 117132776 A CN117132776 A CN 117132776A
Authority
CN
China
Prior art keywords
model
image
encoder
layer
training
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
CN202311174265.9A
Other languages
Chinese (zh)
Inventor
洪波
梁姬慧
季红丽
程国华
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.)
Hangzhou Jianpei Technology Co ltd
Original Assignee
Hangzhou Jianpei 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 Hangzhou Jianpei Technology Co ltd filed Critical Hangzhou Jianpei Technology Co ltd
Priority to CN202311174265.9A priority Critical patent/CN117132776A/en
Publication of CN117132776A publication Critical patent/CN117132776A/en
Pending legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/20Image preprocessing
    • G06V10/26Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • G06N3/0455Auto-encoder networks; Encoder-decoder networks
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/048Activation functions
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/20Image preprocessing
    • G06V10/25Determination of region of interest [ROI] or a volume of interest [VOI]
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/70Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06V10/7747Organisation of the process, e.g. bagging or boosting
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/70Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/806Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/70Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/94Hardware or software architectures specially adapted for image or video understanding
    • G06V10/945User interactive design; Environments; Toolboxes
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02TCLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00Road transport of goods or passengers
    • Y02T10/10Internal combustion engine [ICE] based vehicles
    • Y02T10/40Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Apparatus For Radiation Diagnosis (AREA)

Abstract

The application discloses a multi-organ image segmentation model construction method and a segmentation method, which can complete segmentation of various organs by prompting points or circumscribed rectangular frames only by one model aiming at a wide 3D medical image, and do not need a zero-sample segmentation scheme for retraining a segmentation task of a certain organ. The construction method comprises the following steps: acquiring a pre-training model of image segmentation; adding a control module to the pre-training model to obtain a first model; and freezing an original embedded layer of the pre-training model in the first model, inputting the fine-tuning training set into the first model, and training the first model to obtain a fine-tuned multi-organ image segmentation model. The segmentation method comprises the steps of inputting a medical image to be segmented into a constructed multi-organ image segmentation model; acquiring an operation prompt aiming at a point or a rectangular frame of the medical image to be segmented; and obtaining a segmentation result of the prompt area according to the operation prompt.

Description

Multi-organ image segmentation model construction method and segmentation method
Technical Field
The application relates to the technical field of medical image processing, in particular to a medical image segmentation technology.
Background
Medical image segmentation has made a tremendous contribution in the medical field as an image processing technique. Medical image segmentation is a technique that can extract regions of interest (e.g., tissue and organs, lesions, etc.), and can assist doctors in diagnosing diseases, customizing treatment regimens, etc.
Conventional image segmentation techniques are generally based on algorithms such as threshold segmentation, edge detection or region growing, but in medical image processing, the methods often cannot achieve ideal effects in the face of problems of medical images such as large noise, low contrast, complex structure and the like. Therefore, with the development of deep learning, medical image processing based on neural networks has become a mainstream scheme.
Deep learning models such as convolutional neural networks and the like are widely used for medical image segmentation tasks. These models enable automatic learning of image features by training a large amount of marker data and accurate pixel-level segmentation according to different tasks.
However, labeling of medical image segmentations is time consuming and expensive, and requires specialized imaging physician labeling, as having specialized medical knowledge is adequate for such tasks. Meanwhile, the new segmentation model is re-labeled and trained for images of different tasks and different modes, and the images cannot be reused.
There are two existing approaches to solving the segmentation problem. One is interactive segmentation, which can segment any class of objects, but requires a person to fine tune the mask through iterations. The second is automatic segmentation, which can segment specific objects defined in advance, but the training process requires a large number of manually labeled objects. In summary, neither of these two approaches provides a versatile, fully automated segmentation approach.
Disclosure of Invention
The application aims to overcome the problems in the prior art of medical image segmentation, and provides a multi-organ image segmentation model construction method and a segmentation method, which can complete segmentation of various organs by prompting points or circumscribed rectangular frames only by one model aiming at a wide 3D medical image, and do not need a zero-sample segmentation scheme for retraining a segmentation task of a certain organ.
The application provides a multi-organ image segmentation model construction method, which comprises the following steps:
acquiring a pre-training model of image segmentation, wherein the pre-training model is based on a transducer architecture and comprises an encoder layer encoder, a decoder layer decoder and an embedded layer ebedding output by the encoder layer encoder; the encoder layer encoder is used for mapping an image to be segmented and an input campt to a corresponding feature space and outputting the image to a corresponding embedded layer ebedding, and the campt comprises a point or an external rectangular frame; the encoder layer encoder comprises a Multi-head attention module Multi-head attention and a full connection module;
adding a control module into the pre-training model to obtain a first model, wherein the control module is connected with a Multi-head attention module Multi-head attention in an encoder layer encoder and used for processing the characteristics of the Multi-head attention mechanism module Multi-head attention and fusing the characteristics with a subsequent full-connection module;
collecting a public data set, preprocessing, and obtaining a fine tuning training set, wherein the public data set comprises medical three-dimensional images from multiple suppliers, multiple stages, multiple organs and multiple modes and corresponding segmentation labels;
and freezing the original emmbedding belonging to the pre-training model in the first model, inputting the fine-tuning training set into the first model, and training the first model to obtain a fine-tuned multi-organ image segmentation model.
The application also provides a multi-organ image segmentation method, which is realized based on the multi-organ image segmentation model constructed by the method, and comprises the following steps:
inputting a medical image to be segmented into the multi-organ image segmentation model constructed according to claims 1 to 8;
acquiring an operation prompt aiming at a point or a rectangular frame in a medical image to be segmented, and inputting the operation prompt as a prompt into a multi-organ image segmentation model;
and (3) reasoning the multi-organ image segmentation model, and outputting the segmentation result of the selected organ prompted by the operation.
The technical scheme of the application aims at solving the core problem that in the prior art, special model training and labeling are required for three-dimensional data segmentation tasks of different organs. The scheme is based on the existing pre-training model and sample set, fine adjustment is carried out on the pre-training model, and manual labeling is not needed.
In the fine tuning training, a control module connected with a Multi-head attention module is added on the basis of freezing an original enabling module, an original model can be protected, learned parameters are placed in the control module, and training speed is accelerated while new information is learned.
In practical application, the multi-organ image segmentation model constructed and trained by the scheme does not need to carry out independent special training and labeling on the image segmentation task of a certain organ. Moreover, under the operation prompt of a point or a rectangular frame, the hintable interactive segmentation can be carried out aiming at the hinted organ region, so that the application value of the image segmentation technology in the medical field is greatly improved.
Drawings
FIG. 1 is a flow chart of a method for constructing a multi-organ image segmentation model according to the present application.
Fig. 2 is a schematic diagram of a model constructed by a multi-organ image segmentation model construction method of the present application.
FIG. 3 is a schematic diagram of each transducer structure in the encoder layer encoder in a multi-organ image segmentation model construction method according to the present application.
Fig. 4 is a schematic diagram of a control module in a multi-organ image segmentation model construction method according to the present application.
Fig. 5 is a flowchart of a multi-organ image segmentation method according to the present application.
Fig. 6 is an exemplary diagram of a multi-organ image segmentation method according to the present application.
Description of the embodiments
The application is further described below with reference to the drawings and detailed description.
Example 1
The application relates to a multi-organ image segmentation model construction method, which is shown in fig. 1 as a flow chart of the method, and is shown in fig. 2 as a schematic diagram of a constructed multi-organ image segmentation model, and comprises the following steps of.
Step S101, an acquired pre-training model.
The pre-training model is based on a transducer architecture and comprises an encoder layer encoder, a decoder layer decoder and an embedded layer ebedding output by the encoder layer encoder; the encoder layer encoder is used for mapping the image to be segmented and the input campt to the corresponding feature space and outputting the image to the corresponding embedding layer embedding.
Specifically, the pre-training model adopts various image segmentation pre-training models based on a transducer architecture, such as a SAM segmentation model proposed by Meta. The pre-training model further comprises:
an Image encoder layer Image encoder for mapping an Image to be segmented into an Image feature space;
a prompt encoder layer prompt encoder for mapping input prompt to a characteristic space of the prompt, wherein the prompt comprises a point or an external rectangular frame;
and the encoder layer encoder is used for integrating two embedded layers empedding respectively output by the Image encoder layer Image encoder and the prompt encoder layer template encoder, and then decoding a final segmentation mask from the feature map.
Step S102, adding a control module to the encoder layer encoder of the pre-training model to obtain a first model. The control module is connected with the Multi-head attention module Multi-head attention in the encoder layer encoder and is used for processing the characteristics of the Multi-head attention mechanism module Multi-head attention and fusing the characteristics with the following full-connection module.
In the transducer architecture, the encoder layer encoder includes a Multi-head attention module Multi-head attention and a fully connected module. The encoder layer encoder in this embodiment includes an Image encoder layer Image encoder and a prompt encoder layer sample encoder, which are each formed by connecting n transducer structures in series, as shown in fig. 3, and are schematic diagrams of each transducer structure in the encoder layer encoder.
As shown in fig. 4, the control module adopts a bottleneck structure, and includes a first linear layer, a GeLU activation function layer, and a second linear layer, which are sequentially connected, and is configured to embed the features of the Multi-head attention mechanism module Multi-head attention into a smaller dimension by using the projection mapping of the first linear layer downward by 1/4, and then project back to the original dimension upward through the second linear layer after being processed by the GeLU activation function layer.
The method aims at the Multi-head Attention module to carry out parameter sharing, and then a control module is added to carry out fine adjustment of parameters. In the subsequent structure, a control module is also added to fuse the characteristics of the output of the control module after updating the parameters with the characteristics of the frozen linear module. The method is used for adding a control module to a specific module to finely adjust the SAM architecture under the condition of incompletely changing original parameters so as to improve the capability of the SAM architecture on medical image segmentation.
Step S103, collecting a public data set, and preprocessing to obtain a fine tuning training set. The disclosed dataset includes medical three-dimensional images from multiple suppliers, multiple stages, multiple organs, multiple modalities, and corresponding segmentation annotations.
The disclosed data set comprises a MICCAI 2022 flame abdomen scanning data set, wherein 2300 CT images are contained, the data set is from a 20+ medical institution, and the data set comprises 13 images of parts and segmented labels; amos multi-organ dataset comprising 600 CT and MRI scans comprising labeling of 15 organ segmentations; the brain tumor dataset of BraTs 2021 contained data for four modalities of T1, T1Gd, T2, T2Flair for 660 cases of data; the ACDC heart segmentation data set comprises 200 heart left and right partition labeling results; abdoment CT-1K, having more than 1000 CT scans from 12 medical centers, is a three-dimensional medical image segmentation dataset comprising multi-stage, multi-vendor multi-organ cases.
Specifically, the step of collecting the public data set, preprocessing the public data set and obtaining the fine tuning training set further comprises the following steps.
The hospital image card window width window level in the public dataset is normalized to the interval of 0-255.
Slicing the segmented mask in the z-axis direction to obtain a 2D image, obtaining a 2D slice mask, obtaining a bbox tensor of an external frame obtained by taking a rectangle externally connected with the 2D slice mask, and storing the bbox tensor into a npz file to obtain a fine tuning training set.
Step S104, freezing an original embedded layer embedding belonging to the pre-training model in the first model, inputting a fine-tuning training set into the first model, and training the first model to obtain a fine-tuned multi-organ image segmentation model.
The specific training steps are as follows:
and a data iterator is made and used for iterating and returning the 2D slice image, the 2D slice mask and the corresponding bbox frame tensor from the training set in a batch-by-batch mode and inputting the first model.
Scaling the 2D slice images and the 2D slice mask of the same batch to the same size, and simultaneously adjusting the size of the corresponding bbox frame; the bbox tensor increases the random 20 pixel shift as a perturbation.
The information of the bbox frame is used as prompt template to be encoded and added to a part of a prompt encoder layer template decoder.
The original embedded layer embedding belonging to the pre-training model in the first model is frozen, the data iterator inputs the fine tuning training set into the first model for training, and training learning parameters are input into the control module. In the first model training process, the combination loss of DiceLoss and CELOS is used as the training loss, and the learning rate is 1e-5.
The structure of the control module is mainly that firstly, the first linear layer is used for projecting and mapping downwards to 1/4 of the projection, the Multi-head attention mechanism module Multi-head attention is embedded into a smaller dimension, then the Multi-head attention mechanism module is connected with the GeLU activation function layer for processing and then projected upwards to the original dimension through the second linear layer for fine adjustment of the SAM framework, under the condition that a pre-training model is not trained again on a large scale, the original model can be protected by inserting the control module, the learned parameters are simultaneously put into the control module by the fine adjustment training mode of the scheme, and the training speed is accelerated while new information is learned.
Example 2
As shown in fig. 5, the present application also provides a multi-organ image segmentation method, which is implemented based on the multi-organ image segmentation model constructed by the above method, and includes the following steps:
step S201, inputting the medical image to be segmented into the multi-organ image segmentation model constructed by the multi-organ image segmentation model construction method.
Step S202, acquiring an operation prompt aiming at a point or a rectangular frame in a medical image to be segmented, and inputting the operation prompt into a multi-organ image segmentation model as a prompt.
Step S203, the multi-organ image segmentation model performs reasoning and outputs the segmentation result of the selected organ of the operation prompt.
The step of inputting the medical image to be segmented into the multi-organ image segmentation model further comprises a step of preprocessing, specifically:
slicing and preprocessing a three-dimensional image of a medical image to be segmented into a 2D slice image along the z-axis direction;
and inputting the 2D slice image serving as preprocessed data of the medical image to be segmented into a multi-organ image segmentation model.
The technical scheme of the application aims at solving the core problem that in the prior art, special model training and labeling are required for three-dimensional data segmentation tasks of different organs. The scheme is based on the existing pre-training model and sample set, fine adjustment is carried out on the pre-training model, and manual labeling is not needed.
In the fine tuning training, a control module connected with a Multi-head attention module is added on the basis of freezing an original enabling module, an original model can be protected, learned parameters are placed in the control module, and training speed is accelerated while new information is learned.
In practical application, the multi-organ image segmentation model constructed and trained by the scheme does not need to carry out independent special training and labeling on the image segmentation task of a certain organ. As shown in the example diagram of FIG. 6, under the operation prompt of a point or a rectangular frame, the hintable interactive segmentation can be performed for the hinted organ region, so that the application value of the image segmentation technology in the medical field is greatly improved.

Claims (10)

1. The multi-organ image segmentation model construction method is characterized by comprising the following steps of:
acquiring a pre-training model of image segmentation, wherein the pre-training model is based on a transducer architecture and comprises an encoder layer encoder, a decoder layer decoder and an embedded layer ebedding output by the encoder layer encoder; the encoder layer encoder is used for mapping an image to be segmented and an input campt to a corresponding feature space and outputting the image to a corresponding embedded layer ebedding, and the campt comprises a point or an external rectangular frame; the encoder layer encoder comprises a Multi-head attention module Multi-head attention and a full connection module;
adding a control module into the pre-training model to obtain a first model, wherein the control module is connected with a Multi-head attention module Multi-head attention in an encoder layer encoder and used for processing the characteristics of the Multi-head attention mechanism module Multi-head attention and fusing the characteristics with a subsequent full-connection module;
acquiring a public data set, and preprocessing to obtain a fine tuning training set, wherein the public data set comprises medical three-dimensional images from multiple suppliers, multiple stages, multiple organs and multiple modes and corresponding segmentation labels;
freezing an original embedded layer embedding belonging to a pre-training model in the first model, inputting a fine-tuning training set into the first model, and training the first model to obtain a fine-tuning multi-organ image segmentation model.
2. The method of claim 1, wherein the encoder layer encoder in the pre-training model comprises an Image encoder layer Image encoder and a prompt encoder layer prompt encoder,
an Image encoder layer Image encoder for mapping an Image to be segmented into an Image feature space;
a prompt encoder layer promptness encoder for mapping input promptness to a promptness feature space, wherein the promptness comprises points or circumscribed rectangular boxes;
and the decoding layer decoder is used for integrating two embedded layers enabling the Image encoder layer Image encoder and the prompt encoder layer sample encoder to output respectively, and then decoding a final segmentation mask from the feature map.
3. The method for constructing a multi-organ image segmentation model according to claim 2, wherein the step of acquiring the public dataset and preprocessing the public dataset to obtain the fine tuning training set further comprises the steps of:
normalizing the window width of the hospital image card window in the public data set to a range of 0-255;
slicing the segmented mask in the z-axis direction to obtain a 2D image, obtaining a 2D slice mask, obtaining a bbox tensor of an external frame obtained by taking a rectangle externally connected with the 2D slice mask, and storing the bbox tensor into a npz file to obtain a fine tuning training set.
4. A method for constructing a multi-organ image segmentation model according to claim 3, wherein the step of freezing the original embedding layer embedding belonging to the pre-training model in the first model, inputting the fine-tuning training set into the first model, training the first model, and obtaining the fine-tuned multi-organ image segmentation model further comprises the steps of:
a data iterator is made for iterating from the fine tuning training set back to the 2D slice image, the 2D slice mask and the corresponding bbox frame tensor on a batch-by-batch basis and inputting a first model,
scaling the 2D slice images and the 2D slice masks of the same batch to the same size, and correspondingly adjusting the size of the bbox frame; the bbox tensor increases the random 20 pixel shift as a perturbation;
the information of the bbox frame is used as prompt template to be encoded and added into a part of a prompt encoder layer template decoder;
the original embedded layer embedding belonging to the pre-training model in the first model is frozen, the data iterator inputs the fine tuning training set into the first model for training, and training learning parameters are input into the control module.
5. The method for constructing a multi-organ image segmentation model according to claim 1, wherein in the first model training process, a combination loss of DiceLoss and celos is used as a training loss, and a learning rate is 1e-5.
6. The method for constructing the Multi-organ image segmentation model according to claim 1, wherein the control module adopts a bottleneck structure and comprises a first linear layer, a GeLU activation function layer and a second linear layer which are sequentially connected, and the control module is used for reducing the dimension of the characteristics of the Multi-head attention mechanism module Multi-head attention by using the projection mapping of the first linear layer downwards by 1/4, and projecting the Multi-head attention mechanism module Multi-head attention back to the original dimension through the second linear layer after being processed by the GeLU activation function layer.
7. The method for constructing a multi-organ image segmentation model according to claim 1, wherein the pre-training model is a SAM model.
8. The method of claim 1, wherein the public dataset is a three-dimensional medical image segmentation dataset comprising multi-stage, multi-vendor, multi-organ cases, including MICCAI 2022 flere abdominal scan dataset, amos multi-organ dataset, braTs 2021 brain tumor dataset, ACDC heart segmentation dataset, abdoment-1K dataset.
9. A method of multi-organ image segmentation, the method comprising the steps of:
inputting a medical image to be segmented into the multi-organ image segmentation model constructed according to claims 1 to 8;
acquiring an operation prompt aiming at a point or a rectangular frame in a medical image to be segmented, and inputting the operation prompt as a prompt into a multi-organ image segmentation model;
and (3) reasoning the multi-organ image segmentation model, and outputting the segmentation result of the selected organ prompted by the operation.
10. The method of claim 9, wherein the step of inputting the medical image to be segmented into the multi-organ image segmentation model constructed in claims 1 to 8 further comprises:
slicing and preprocessing a three-dimensional image of a medical image to be segmented into a 2D slice image along the z-axis direction;
inputting the 2D slice image as preprocessed data of the medical image to be segmented into the multi-organ image segmentation model constructed according to the claims 1 to 8.
CN202311174265.9A 2023-09-12 2023-09-12 A multi-organ image segmentation model construction method and segmentation method Pending CN117132776A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311174265.9A CN117132776A (en) 2023-09-12 2023-09-12 A multi-organ image segmentation model construction method and segmentation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311174265.9A CN117132776A (en) 2023-09-12 2023-09-12 A multi-organ image segmentation model construction method and segmentation method

Publications (1)

Publication Number Publication Date
CN117132776A true CN117132776A (en) 2023-11-28

Family

ID=88859925

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311174265.9A Pending CN117132776A (en) 2023-09-12 2023-09-12 A multi-organ image segmentation model construction method and segmentation method

Country Status (1)

Country Link
CN (1) CN117132776A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117788981A (en) * 2024-02-02 2024-03-29 深圳须弥云图空间科技有限公司 Zero-shot image segmentation model training method and device based on multimodality
CN118072378A (en) * 2024-03-11 2024-05-24 珠海全一科技有限公司 Dynamic decision image segmentation method based on SAM basic model
CN118447038A (en) * 2024-04-17 2024-08-06 中国科学院自动化研究所 Interactive image segmentation method and device, electronic equipment and storage medium

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190223725A1 (en) * 2018-01-25 2019-07-25 Siemens Healthcare Gmbh Machine Learning-based Segmentation for Cardiac Medical Imaging
CN110136828A (en) * 2019-05-16 2019-08-16 杭州健培科技有限公司 A method of medical image multitask auxiliary diagnosis is realized based on deep learning
CN113808146A (en) * 2021-10-18 2021-12-17 山东大学 Medical image multi-organ segmentation method and system
CN115018865A (en) * 2022-06-30 2022-09-06 西安理工大学 Medical image segmentation method based on transfer learning
CN116309650A (en) * 2023-05-22 2023-06-23 湖南大学 Medical image segmentation method and system based on dual-branch embedded attention mechanism
GB202307321D0 (en) * 2023-05-17 2023-06-28 Samsung Electronics Co Ltd Method and apparatus for fine-tuning foundation models

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190223725A1 (en) * 2018-01-25 2019-07-25 Siemens Healthcare Gmbh Machine Learning-based Segmentation for Cardiac Medical Imaging
CN110136828A (en) * 2019-05-16 2019-08-16 杭州健培科技有限公司 A method of medical image multitask auxiliary diagnosis is realized based on deep learning
CN113808146A (en) * 2021-10-18 2021-12-17 山东大学 Medical image multi-organ segmentation method and system
CN115018865A (en) * 2022-06-30 2022-09-06 西安理工大学 Medical image segmentation method based on transfer learning
GB202307321D0 (en) * 2023-05-17 2023-06-28 Samsung Electronics Co Ltd Method and apparatus for fine-tuning foundation models
CN116309650A (en) * 2023-05-22 2023-06-23 湖南大学 Medical image segmentation method and system based on dual-branch embedded attention mechanism

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
JUNDE WU等: "Medical SAM Adapter: Adapting Segment Anything Model for Medical Image Segmentation", MEDICAL SAM ADAPTER: ADAPTING SEGMENT ANYTHING MODEL FOR MEDICAL IMAGE SEGMENTATION, 13 May 2023 (2023-05-13), pages 1 - 13 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117788981A (en) * 2024-02-02 2024-03-29 深圳须弥云图空间科技有限公司 Zero-shot image segmentation model training method and device based on multimodality
CN118072378A (en) * 2024-03-11 2024-05-24 珠海全一科技有限公司 Dynamic decision image segmentation method based on SAM basic model
CN118447038A (en) * 2024-04-17 2024-08-06 中国科学院自动化研究所 Interactive image segmentation method and device, electronic equipment and storage medium

Similar Documents

Publication Publication Date Title
CN112465827B (en) Contour perception multi-organ segmentation network construction method based on class-by-class convolution operation
CN113674253B (en) Automatic segmentation method for rectal cancer CT image based on U-transducer
JP7707299B2 (en) Automatic liver CT segmentation method based on deep shape learning
CN114693933B (en) Medical image segmentation device based on generative adversarial network and multi-scale feature fusion
CN117132776A (en) A multi-organ image segmentation model construction method and segmentation method
CN110310287B (en) Automatic organ-at-risk delineation method, equipment and storage medium based on neural network
Zhang et al. Automatic segmentation of the cardiac MR images based on nested fully convolutional dense network with dilated convolution
WO2025050542A1 (en) Diffusion model and generative adversarial network-based image segmentation method and apparatus
CN113554669A (en) Unet network brain tumor MRI image segmentation method for improving attention module
CN113706486B (en) Pancreatic tumor image segmentation method based on dense connection network migration learning
WO2021203795A1 (en) Pancreas ct automatic segmentation method based on saliency dense connection expansion convolutional network
CN111179237A (en) A kind of liver and liver tumor image segmentation method and device
CN117710681A (en) Semi-supervised medical image segmentation method based on data enhancement strategy
Zeng et al. Liver segmentation in magnetic resonance imaging via mean shape fitting with fully convolutional neural networks
CN113160229A (en) Pancreas segmentation method and device based on hierarchical supervision cascade pyramid network
CN107516314B (en) Medical image hyper-voxel segmentation method and device
CN119992083A (en) Abdominal CT multi-organ segmentation method based on LoRA vertical domain MedSAM large model
CN118762177A (en) An automatic segmentation method for pancreatic tumors based on CT images
CN116402831A (en) Partially supervised method and device for automatic segmentation of multiple organs in abdominal CT sequence images
CN120525806A (en) Interactive recognition and measurement method and device based on AI non-enhanced CT images
CN120072213A (en) Semi-supervised medical image segmentation method based on cross-image learning and shape fusion
Gibou et al. Partial differential equations-based segmentation for radiotherapy treatment planning
CN119379710A (en) A semi-supervised medical image segmentation method based on mutual correction of suspicious pixels
CN113538451B (en) Method and device for segmenting magnetic resonance image of deep vein thrombosis, electronic equipment and storage medium
Bransby et al. POLYCORE: Polygon-based contour refinement for improved Intravascular Ultrasound Segmentation

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