CN111784682A - Network architecture, program carrier and workstation for automatic processing of images - Google Patents

Network architecture, program carrier and workstation for automatic processing of images Download PDF

Info

Publication number
CN111784682A
CN111784682A CN202010663227.XA CN202010663227A CN111784682A CN 111784682 A CN111784682 A CN 111784682A CN 202010663227 A CN202010663227 A CN 202010663227A CN 111784682 A CN111784682 A CN 111784682A
Authority
CN
China
Prior art keywords
images
image
path
network architecture
encoding
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
CN202010663227.XA
Other languages
Chinese (zh)
Inventor
王少彬
陈颀
陈宇
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.)
Beijing Yizhiying Technology Co ltd
Original Assignee
Beijing Yizhiying 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 Beijing Yizhiying Technology Co ltd filed Critical Beijing Yizhiying Technology Co ltd
Priority to CN202010663227.XA priority Critical patent/CN111784682A/en
Publication of CN111784682A publication Critical patent/CN111784682A/en
Priority to PCT/CN2021/105538 priority patent/WO2022007957A1/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10081—Computed x-ray tomography [CT]
    • 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
    • G06T2207/20092—Interactive image processing based on input by user
    • G06T2207/20104—Interactive definition of region of interest [ROI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30096—Tumor; Lesion

Abstract

A network architecture (1) for automatic processing of images is disclosed, comprising: an input module (11) for inputting an image to be processed; an encoding path (12) configured to perform feature extraction on an input image using a dual path network; a decoding path (14) configured to establish a connection with the encoding path (12); a central module (13) configured for a transition from an encoding path (12) to a decoding path (14) to refine high-dimensional image features; and an output module (15) configured to output the image processing result from the decoding path (14); wherein the decoding path (14) is configured to perform a decoding operation on the respective encoding results of the encoding path (12) and the image feature refinement results of the central module (13) using a decoding path of the Unet architecture. A corresponding method, a corresponding program carrier and a corresponding workstation are also disclosed. According to the invention, better image segmentation and classification can be carried out.

Description

Network architecture, program carrier and workstation for automatic processing of images
Technical Field
The invention relates to a network architecture for automatic processing of images, a corresponding computer-readable program carrier and a corresponding workstation.
Background
There are many occasions when it is necessary to process an image, for example by segmenting the image to identify or automatically contour a particular object in the image.
With the development of modern medicine, more and more diseases are diagnosed and treated by means of medical images, such as tumor radiotherapy.
Cervical cancer is now the second most common cancer in women between the ages of 15 and 44 worldwide. Intensive-modulation therapy (IMRT) has become the radiation therapy of choice for the treatment of cervical cancer. The effectiveness of IMRT depends on the accuracy of the delineation of the Clinical Target Volume (CTV) and the organ-at-risk (OAR). The contouring of clinical target areas and organs at risk is currently accomplished by radiation therapy oncologists through laborious and cumbersome manual delineations. Contouring is very time consuming and depends very much on the experience of the radiation therapist. Despite the availability of standard guidelines, the differences in the experience of radiation therapy oncologists remain one of the major challenges in planning radiation therapy. Thus, the manual workload of a radiation therapy oncologist can be greatly reduced if the anatomy can be automatically segmented in a reasonable time.
Traditional automatic segmentation methods are derived from statistical models or atlas-based models. However, both of these approaches have limitations. In particular, the organs where the cervical cancer is located are complex, and the boundaries of the CT image are not clear. Therefore, current automated segmentation methods perform poorly in the contouring of CTVs, various complex tissues and organs may be confused with the boundaries of CTVs, and potential spread of tumor tissue or subclinical disease in CTVs may not be detectable in CT images.
Thus, there is a great need for improvement, particularly for cancers such as cervical cancer, for which automatic contouring is currently difficult to achieve well.
Disclosure of Invention
The object of the invention is achieved by an improved network architecture for automated processing of images, a corresponding computer-readable program carrier and a corresponding workstation.
According to a first aspect of the present invention, there is provided a network architecture for automated processing of images, the network architecture comprising: the input module is used for inputting an image to be processed; an encoding path configured to perform feature extraction on an input image using a dual path network; a decoding path configured to establish a connection with the encoding path; a central module configured for a transition from an encoding path to a decoding path to refine high-dimensional image features; and an output module configured to output the image processing result from the decoding path; wherein the decoding path is configured to perform a decoding operation on the respective encoding results of the encoding paths and the image feature refinement result of the central module using a decoding path of the Unet architecture.
According to an alternative embodiment of the invention, the dual path network comprises a series of concatenated micro-blocks, which are accordingly embedded in the decoder of the decoding path.
According to an alternative embodiment of the invention, the encoding path comprises 5 encoders and the decoding path comprises 4 decoders, each decoder using a respective micro-block in the encoding path.
According to an alternative embodiment of the invention, the central module is configured to implement: conv (3 × 3) + BN + ReLu; and the decoder is configured to implement:
Figure BDA0002579387240000021
wherein Conv (3 × 3) is a 3 × 3 convolution operation, BN is a batch normalization operation, ReLU is a linear rectification function,
Figure BDA0002579387240000022
representing bilinear upsampling.
According to an alternative embodiment of the present invention, the input module receives an image or images, and the output module outputs an image or images; and the output module is configured to implement:
Figure BDA0002579387240000023
wherein the content of the first and second substances,
Figure BDA0002579387240000024
representing a Sigmoid activation function.
According to an optional embodiment of the present invention, the image is a computed tomography image, and the plurality of images are a plurality of computed tomography images adjacent to each other in front and back; and/or the network architecture is configured for automatic segmentation processing of images.
According to an optional embodiment of the invention, the input module receives an image or a plurality of images, and the output module outputs a classification result; and the output module is configured to implement:
Figure BDA0002579387240000025
Figure BDA0002579387240000026
wherein the content of the first and second substances,
Figure BDA0002579387240000027
representing a fully connected layer.
According to an optional embodiment of the present invention, a loss function is configured at an end point of the network architecture, and the loss function is a cross-entropy loss function or a weight adaptive loss function:
Figure BDA0002579387240000031
Figure BDA0002579387240000032
wherein the content of the first and second substances,
Figure BDA0002579387240000033
Figure BDA0002579387240000034
p is Sigmoid activation function output; y is a label graph, which takes the value 0 or 1; i is an input image; n is the number of total segmentation targets; p and y both have the shape I x N; pnThe counted volume size of the target region of each category is the proportion of the size of the whole image volume.
According to an alternative embodiment of the invention, when the network architecture is configured to output a plurality of segmented images, a formula is used as the loss function.
According to an alternative embodiment of the invention, the network architecture is configured for automatic segmentation or classification of images for tumor radiotherapy, in particular for cervical cancer radiotherapy.
According to a second aspect of the invention, there is provided a method of processing an image using the network architecture, comprising the steps of: inputting an image through an input module; processing the image through the encoding path, the central module and the decoding path; and outputting the image processing result through the output module.
According to a third aspect of the invention, there is provided a computer readable program carrier storing program instructions for implementing the method when executed by a processor.
According to a fourth aspect of the invention, there is provided a workstation configured to comprise the computer readable program carrier.
According to an alternative embodiment of the invention, the workstation is configured as a physician workstation for automatic segmentation or classification of medical images.
According to the invention, not only can the performance of image processing be improved, but also the network can be established more quickly.
Drawings
The principles, features and advantages of the present invention may be better understood by describing the invention in more detail below with reference to the accompanying drawings. The drawings comprise:
FIG. 1 shows a simplified block diagram of a network architecture for automatic segmentation or classification of images according to an exemplary embodiment of the present invention.
Fig. 2 schematically shows a simplified block diagram of a network architecture according to an exemplary embodiment of the present invention.
Fig. 3 schematically shows a simplified block diagram of a network architecture for automatic segmentation of images according to another exemplary embodiment of the present invention.
Fig. 4 schematically shows a simplified block diagram of a network architecture for automatic segmentation of images according to yet another exemplary embodiment of the present invention.
Fig. 5 schematically shows a simplified block diagram for identifying which regions of interest are in an image according to an exemplary embodiment of the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantageous effects of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings and exemplary embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of the invention.
FIG. 1 shows a simplified block diagram of a network architecture for automatic segmentation or classification of images according to an exemplary embodiment of the present invention.
As shown in fig. 1, the network architecture 1 may mainly include an input module 11, an encoding path 12, a central module 13, a decoding path 14 and an output module 15, wherein the input module 11 is configured to receive an image, for example, a CT image, the encoding path 12 is configured to use a Dual Path Network (DPN) to perform feature extraction on the input image, the decoding path 14 is associated with the encoding path 12 to correspondingly decode an encoding result of the encoding path 12, the central module 13 is configured to transition from the encoding path 12 to the decoding path 14 to refine high-dimensional image features, the output module 15 is configured to output an output result that can be used for automatic contour delineation, and the decoding path 14 is configured to decode using a decoding path of a uet architecture.
Dual path networks, in turn, include residual network (ResNet) paths, which support reuse of features, and dense convolutional network (densneet) paths, which support exploration of new features, thus combining the advantages of both.
The dual path network includes a series of Micro-blocks (Micro-blocks) connected in series. The micro-tiles are the core of the DPN. By using a dual path network as the encoder portion of the network architecture of the present invention, better feature extraction capabilities can be achieved.
According to an exemplary embodiment of the present invention, in order to make the decoder section have the same performance of recovering the abstract features, the micro-blocks are embedded in the decoder section to replace the standard convolution operation.
Fig. 2 schematically shows a simplified block diagram of a network architecture 1 according to an exemplary embodiment of the present invention.
As shown in fig. 2, the single CT image 16 input to the encoding path 12 has pixels of 512 × 512, and 1 × 512 represents a CT image having pixels of 512 × 512. The CT image 16 is subjected to a series of encoder processes in the encoding path 12 to extract features of the image. The final encoded result of the encoding path 12 is input to the central module 13, and then processed by the central module 13 and output to the decoding path 14, so as to realize the transition from the encoding path 12 to the decoding path 14. It can also be seen from fig. 2 that the decoding path 14 is also associated with the encoding path 12 as schematically indicated by arrow 17.
In the exemplary embodiment shown in fig. 2, the encoding path 12 encodes the image 16 in a series by 5 encoders/micro-blocks 18, while the decoding path 14 includes 4 decoders 19, each decoder 19 using a respective micro-block/encoder 18 in the encoding path 12.
In fig. 2, exemplary operations within some of the modules are also represented schematically by symbols. Specifically, darker arrows indicate Conv (3 × 3) + BN + ReLu, where Conv (3 × 3) is a 3 × 3 convolution operation, BN (Batch Normalization) is a Batch Normalization operation, and ReLu (rectified Linear unit) is a Linear rectification function. Lighter arrows indicate tiles.
Figure BDA0002579387240000051
Representing the connection relationship.
Figure BDA0002579387240000052
Representing a bi-linear up-sampling of the samples,
Figure BDA0002579387240000053
representing a Sigmoid activation function.
According to an exemplary embodiment of the present invention, as shown in fig. 2, the central module 13 may be configured to: conv (3 × 3) + BN + ReLu.
In the exemplary embodiment shown in fig. 2, the image processed by the decoding path 14 is output to the output module 15, and the output module 15 also outputs an automatically segmented 512 × 512 image 20.
According to an exemplary embodiment of the present invention, as shown in fig. 2, the output module 15 may be configured to:
Figure BDA0002579387240000061
wherein the Sigmoid activates the function
Figure BDA0002579387240000062
For outputting a probability value that each pixel belongs to a region of interest (ROI).
According to an exemplary embodiment of the present invention, as shown in fig. 2, the decoder 19 may be configured to:
Figure BDA0002579387240000063
the present exemplary embodiment can be used for collaborative planning of tumor radiotherapy, particularly radiotherapy in which the tumor site organs are complex, such as cervical cancer, and more particularly for automatic contouring of clinical target areas and organs-at-risk to generate intermediate results usable by tumor radiotherapy physicians.
Fig. 3 schematically shows a simplified block diagram of a network architecture for automatic segmentation of images according to another exemplary embodiment of the present invention.
The main difference between fig. 3 and fig. 2 is that three CT images 16 adjacent to each other in the front and back are input through the input module 11, so that the neural network obtains image information in the front and back, and is a quasi or 3D-like image. Therefore, the segmentation results of the front layer and the rear layer are smoother, and the sudden change of the segmentation results is avoided.
Of course, it will be understood by those skilled in the art that the input CT images need not be three images, but may be any suitable number of images.
The output is also an automatically segmented image 20. The rest is similar to fig. 2 and will not be described again here for the sake of clarity.
Fig. 4 schematically shows a simplified block diagram of a network architecture for automatic segmentation of images according to yet another exemplary embodiment of the present invention.
Fig. 4 differs from fig. 3 mainly in that the output is a multi-channel output. The output of multiple channels shows that the network not only can perform image segmentation of 2 classes, but also has the function of multi-class classification. Here, for example, cervical cancer, 8 ROI segmentation results can be simultaneously output. The 8 ROIs include 7 organs at risk and 1 clinical target area, specifically: 1. bladder, 2. marrow, 3. left femoral head, 4. right femoral head, 5. rectum, 6. small intestine, 7. spinal cord, 8. cervical carcinoma clinical target area. However, multi-channel outputs for other tumours are also supported, and of course other numbers of channels are also supported.
The multi-channel output representation outputs a plurality of segmented images, each segmented image corresponding to a region of interest.
The others are similar to fig. 3 and are not repeated here for clarity.
Fig. 5 schematically shows a simplified block diagram for identifying which regions of interest are in an image according to an exemplary embodiment of the present invention.
The main difference between fig. 5 and fig. 4 is that the last output part of the output module 15 is replaced by a fully Connected Layer (FC), so that the split network becomes a classified network. The network can determine which ROI are included in the input CT image, but does not determine the position of each ROIAnd (4) judging, namely not performing specific segmentation. In particular, here is the Sigmoid activation function shown in fig. 2-4
Figure BDA0002579387240000071
Replaced by a fully-connected layer
Figure BDA0002579387240000072
Other similar parts are not described in detail.
For better evaluation, training and updating of the network, a loss function is usually provided at the end point of the network, taking as input the predicted value of the network and the prepared gold standard, calculating the loss (deviation or error) by the loss function, and then propagating back to the corresponding, e.g. all, layers in the network to update the corresponding weight values, e.g. ownership weight values, in the network so that the network predicted value gets closer to the gold standard. The network only finally determines, for example when the losses have fallen to an acceptable level, for example a predetermined level, and the optimal weight values are also updated, at which point the training can be ended.
Thus, the loss function not only affects the training process of the network, but also directly affects the final performance of the network.
According to an exemplary embodiment of the present invention, the loss function is a cross-entropy loss function (1) or a weight adaptive loss function (2):
Figure BDA0002579387240000073
Figure BDA0002579387240000074
wherein the content of the first and second substances,
Figure BDA0002579387240000075
Figure BDA0002579387240000076
p is Sigmoid activation function output; y is a label Map (Ground Truth Map) with a value of 0 or 1; i is an input image; n is the total number of segmented objects. p and y both have the shape I x N; pnIs the ratio of the counted volume size of each category of target region, e.g., the organ at risk, to the size of the overall image volume.
Preferably, for the multi-channel split network exemplarily shown in fig. 4, equation (2) is used as the loss function. The multi-channel segmentation network needs to segment a plurality of interested areas, such as a plurality of organs at risk and clinical target areas, if the size difference of the volumes of the plurality of interested areas is large, a better segmentation effect is difficult to obtain by using a cross entropy loss function, and a very good segmentation effect can be obtained by using a weight adaptive loss function.
For other split networks, equation (1) may be used as the loss function.
It will be appreciated that the above description has been made primarily with respect to CT images, but it will be apparent that the images may also be X-ray imaging images, Magnetic Resonance Imaging (MRI) images, Single Photon Emission Computed Tomography (SPECT) images, positron emission computed tomography (PET) images, ultrasound imaging images, and the like.
It will be appreciated by those skilled in the art that the present invention provides, in general, a neural network architecture for automated processing, e.g., automated segmentation or classification, of images, which may be used in the medical field, in particular to assist in segmenting or classifying CT images. In particular, the neural network architecture of the present invention is adapted to facilitate operations involved in automated contouring of clinical targets and organs-at-risk for radiation therapy of tumors (especially cervical cancer), and the physician can then perform subsequent operations, such as review, based on the processing results, such as automated contouring results, and can modify them as necessary for subsequent therapeutic purposes.
Therefore, the network architecture of the present invention is not directly used for therapeutic purposes when automatically processing CT images, but generates intermediate results, mainly aiming at improving the working efficiency of doctors.
Furthermore, it will be appreciated by those skilled in the art that the present invention also provides a computer readable program carrier storing program instructions for implementing the functions of the network architecture described above for image processing when executed by a processor. For example, the computer readable program carrier may be configured as a flash memory or the like.
The invention further provides a workstation configured to include the computer readable program carrier. The workstation may be configured in particular as a doctor workstation for use by a doctor.
Although specific embodiments of the invention have been described herein in detail, they have been presented for purposes of illustration only and are not to be construed as limiting the scope of the invention. Various substitutions, alterations, and modifications may be devised without departing from the spirit and scope of the present invention.

Claims (14)

1. A network architecture (1) for automatically processing images, the network architecture (1) comprising:
an input module (11) for inputting an image to be processed;
an encoding path (12) configured to perform feature extraction on an input image using a dual path network;
a decoding path (14) configured to establish a connection with the encoding path (12);
a central module (13) configured for a transition from an encoding path (12) to a decoding path (14) to refine high-dimensional image features; and
an output module (15) configured to output image processing results from the decoding path (14);
wherein the decoding path (14) is configured to perform a decoding operation on the respective encoding results of the encoding path (12) and the image feature refinement results of the central module (13) using a decoding path of the Unet architecture.
2. The network architecture (1) according to claim 1,
the dual path network comprises a series of concatenated micro-blocks (18), the micro-blocks (18) being embedded in a decoder of the decoding path (14) accordingly.
3. Network architecture (1) according to claim 1 or 2,
the encoding path (12) comprises 5 encoders and the decoding path (14) comprises 4 decoders (19), each decoder (19) using a respective micro-block in the encoding path (12).
4. The network architecture (1) of claim 3,
the central module (13) is configured to implement: conv (3 × 3) + BN + ReLu; and
the decoder (19) is configured to implement: microblock +
Figure FDA0002579387230000011
Wherein Conv (3 × 3) is a 3 × 3 convolution operation, BN is a batch normalization operation, ReLU is a linear rectification function,
Figure FDA0002579387230000012
representing bilinear upsampling.
5. The network architecture (1) of claim 4,
the input module (11) receives one image or a plurality of images, and the output module (15) outputs one image or a plurality of images; and
the output module (15) is configured to implement:
Figure FDA0002579387230000021
wherein the content of the first and second substances,
Figure FDA0002579387230000022
representing a Sigmoid activation function.
6. The network architecture (1) of claim 5,
the images are computed tomography images, X-ray imaging images, magnetic resonance imaging images, single photon emission computed tomography images, positron emission computed tomography images and ultrasonic imaging images, and the images are adjacent images in the front and back; and/or
The network architecture (1) is configured for automatic segmentation processing of images.
7. The network architecture (1) of claim 4,
the input module (11) receives one image or a plurality of images, and the output module (15) outputs a classification result; and
the output module (15) is configured to implement:
Figure FDA0002579387230000023
wherein the content of the first and second substances,
Figure FDA0002579387230000024
representing a fully connected layer.
8. The network architecture (1) of claim 6,
configuring a loss function at an end point of the network architecture (1), wherein the loss function is a cross-entropy loss function (1) or a weight adaptive loss function (2):
Figure FDA0002579387230000025
Figure FDA0002579387230000026
wherein the content of the first and second substances,
Figure FDA0002579387230000027
Figure FDA0002579387230000031
p is Sigmoid activation function output; y is a label graph, which takes the value 0 or 1; i is an input image; n is the number of total segmentation targets; p and y both have the shape I x N; pnThe counted volume size of the target region of each category is the proportion of the size of the whole image volume.
9. The network architecture (1) of claim 8,
when the network architecture (1) is configured to output a plurality of divided images, equation (2) is used as a loss function.
10. The network architecture (1) according to any one of claims 1-9,
the network architecture (1) is configured for automatic segmentation or classification of images for tumor radiotherapy, in particular for cervical cancer radiotherapy.
11. A method of processing an image using the network architecture (1) of any of claims 1-10, comprising the steps of:
inputting an image through an input module (11);
processing the image by means of an encoding path (12), a central module (13) and a decoding path (14); and
the image processing result is output through an output module (15).
12. A computer readable program carrier storing program instructions for implementing the method according to claim 12 when executed by a processor.
13. A workstation configured to include the computer readable program carrier of claim 12.
14. The workstation according to claim 13, wherein the workstation is configured as a physician workstation for automatic segmentation or classification of medical images.
CN202010663227.XA 2020-07-10 2020-07-10 Network architecture, program carrier and workstation for automatic processing of images Pending CN111784682A (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202010663227.XA CN111784682A (en) 2020-07-10 2020-07-10 Network architecture, program carrier and workstation for automatic processing of images
PCT/CN2021/105538 WO2022007957A1 (en) 2020-07-10 2021-07-09 Network architecture for automatically processing images, program carrier, and workstation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010663227.XA CN111784682A (en) 2020-07-10 2020-07-10 Network architecture, program carrier and workstation for automatic processing of images

Publications (1)

Publication Number Publication Date
CN111784682A true CN111784682A (en) 2020-10-16

Family

ID=72768151

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010663227.XA Pending CN111784682A (en) 2020-07-10 2020-07-10 Network architecture, program carrier and workstation for automatic processing of images

Country Status (2)

Country Link
CN (1) CN111784682A (en)
WO (1) WO2022007957A1 (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112734762A (en) * 2020-12-31 2021-04-30 西华师范大学 Dual-path UNet network tumor segmentation method based on covariance self-attention mechanism
CN113378933A (en) * 2021-06-11 2021-09-10 合肥合滨智能机器人有限公司 Thyroid ultrasound image classification and segmentation network, training method, device and medium
WO2022007957A1 (en) * 2020-07-10 2022-01-13 北京医智影科技有限公司 Network architecture for automatically processing images, program carrier, and workstation
CN117197156A (en) * 2022-10-21 2023-12-08 南华大学 Lesion segmentation method and system based on double decoders UNet and Transformer

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110059772A (en) * 2019-05-14 2019-07-26 温州大学 Remote sensing images semantic segmentation method based on migration VGG network
US20190355102A1 (en) * 2018-05-15 2019-11-21 Adobe Inc. Digital Image Completion by Learning Generation and Patch Matching Jointly

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9668699B2 (en) * 2013-10-17 2017-06-06 Siemens Healthcare Gmbh Method and system for anatomical object detection using marginal space deep neural networks
CN109598727B (en) * 2018-11-28 2021-09-14 北京工业大学 CT image lung parenchyma three-dimensional semantic segmentation method based on deep neural network
CN111242288B (en) * 2020-01-16 2023-06-27 浙江工业大学 Multi-scale parallel deep neural network model construction method for lesion image segmentation
CN111369574B (en) * 2020-03-11 2023-05-16 合肥凯碧尔高新技术有限公司 Thoracic organ segmentation method and device
CN111784682A (en) * 2020-07-10 2020-10-16 北京医智影科技有限公司 Network architecture, program carrier and workstation for automatic processing of images

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190355102A1 (en) * 2018-05-15 2019-11-21 Adobe Inc. Digital Image Completion by Learning Generation and Patch Matching Jointly
CN110059772A (en) * 2019-05-14 2019-07-26 温州大学 Remote sensing images semantic segmentation method based on migration VGG network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
崔昊;: "基于深度学习的高分辨率遥感图像海陆分割方法", 软件导刊, no. 03 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022007957A1 (en) * 2020-07-10 2022-01-13 北京医智影科技有限公司 Network architecture for automatically processing images, program carrier, and workstation
CN112734762A (en) * 2020-12-31 2021-04-30 西华师范大学 Dual-path UNet network tumor segmentation method based on covariance self-attention mechanism
CN113378933A (en) * 2021-06-11 2021-09-10 合肥合滨智能机器人有限公司 Thyroid ultrasound image classification and segmentation network, training method, device and medium
CN117197156A (en) * 2022-10-21 2023-12-08 南华大学 Lesion segmentation method and system based on double decoders UNet and Transformer
CN117197156B (en) * 2022-10-21 2024-04-02 南华大学 Lesion segmentation method and system based on double decoders UNet and Transformer

Also Published As

Publication number Publication date
WO2022007957A1 (en) 2022-01-13

Similar Documents

Publication Publication Date Title
Fu et al. A review of deep learning based methods for medical image multi-organ segmentation
CN111784682A (en) Network architecture, program carrier and workstation for automatic processing of images
Lambert et al. Segthor: Segmentation of thoracic organs at risk in ct images
Dong et al. Automatic multiorgan segmentation in thorax CT images using U‐net‐GAN
Men et al. Deep deconvolutional neural network for target segmentation of nasopharyngeal cancer in planning computed tomography images
CN114037726A (en) Atlas-based segmentation using deep learning
Zhang et al. Deep learning initialized and gradient enhanced level-set based segmentation for liver tumor from CT images
Lei et al. Deep learning in multi-organ segmentation
CN112770811A (en) Method and system for radiation therapy treatment planning using a deep learning engine
US9098912B2 (en) Method, system and computer readable medium for automatic segmentation of a medical image
CN110619639A (en) Method for segmenting radiotherapy image by combining deep neural network and probability map model
Shi et al. Deep learning empowered volume delineation of whole-body organs-at-risk for accelerated radiotherapy
JP2014502531A (en) System and method for automatically generating an initial radiation treatment plan
Kumar et al. Dual feature extraction based convolutional neural network classifier for magnetic resonance imaging tumor detection using U-Net and three-dimensional convolutional neural network
Schreier et al. A full-image deep segmenter for CT images in breast cancer radiotherapy treatment
CN113436173B (en) Abdominal multi-organ segmentation modeling and segmentation method and system based on edge perception
US20230086070A1 (en) Image processing methods and systems
US20230290480A1 (en) Systems and methods for clinical target contouring in radiotherapy
Luan et al. Adaptive attention convolutional neural network for liver tumor segmentation
Vandewinckele et al. Segmentation of head-and-neck organs-at-risk in longitudinal CT scans combining deformable registrations and convolutional neural networks
Francis et al. ThoraxNet: a 3D U-Net based two-stage framework for OAR segmentation on thoracic CT images
Kawahara et al. Stepwise deep neural network (stepwise-net) for head and neck auto-segmentation on CT images
Lei et al. Deep learning architecture design for multi-organ segmentation
Singhrao et al. A generative adversarial network‐based (GAN‐based) architecture for automatic fiducial marker detection in prostate MRI‐only radiotherapy simulation images
Chen et al. A multiple organ segmentation system for CT image series using Attention-LSTM fused U-Net

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