CN115100252A - Four-dimensional CT registration method and device for pancreatic region - Google Patents
Four-dimensional CT registration method and device for pancreatic region Download PDFInfo
- Publication number
- CN115100252A CN115100252A CN202210589627.XA CN202210589627A CN115100252A CN 115100252 A CN115100252 A CN 115100252A CN 202210589627 A CN202210589627 A CN 202210589627A CN 115100252 A CN115100252 A CN 115100252A
- Authority
- CN
- China
- Prior art keywords
- image
- registration
- pancreas
- fixed
- dimensional
- 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
- 238000000034 method Methods 0.000 title claims abstract description 48
- 210000000496 pancreas Anatomy 0.000 claims abstract description 117
- 230000011218 segmentation Effects 0.000 claims abstract description 47
- 238000004364 calculation method Methods 0.000 claims abstract description 27
- 238000013528 artificial neural network Methods 0.000 claims abstract description 26
- 230000008569 process Effects 0.000 claims abstract description 18
- 210000001015 abdomen Anatomy 0.000 claims abstract description 11
- 230000006870 function Effects 0.000 claims description 21
- 230000003187 abdominal effect Effects 0.000 claims description 11
- 238000005457 optimization Methods 0.000 claims description 10
- 230000009466 transformation Effects 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 239000000203 mixture Substances 0.000 claims description 5
- 230000001154 acute effect Effects 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- OAICVXFJPJFONN-UHFFFAOYSA-N Phosphorus Chemical compound [P] OAICVXFJPJFONN-UHFFFAOYSA-N 0.000 claims description 2
- 238000013527 convolutional neural network Methods 0.000 claims description 2
- 240000007594 Oryza sativa Species 0.000 claims 1
- 235000007164 Oryza sativa Nutrition 0.000 claims 1
- 235000009566 rice Nutrition 0.000 claims 1
- 238000002591 computed tomography Methods 0.000 description 123
- 238000001959 radiotherapy Methods 0.000 description 6
- 230000000241 respiratory effect Effects 0.000 description 4
- 206010061902 Pancreatic neoplasm Diseases 0.000 description 3
- 210000004072 lung Anatomy 0.000 description 3
- 201000002528 pancreatic cancer Diseases 0.000 description 3
- 238000011282 treatment Methods 0.000 description 3
- 206010028980 Neoplasm Diseases 0.000 description 2
- 206010056342 Pulmonary mass Diseases 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000000988 bone and bone Anatomy 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 208000015486 malignant pancreatic neoplasm Diseases 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000000082 organ preservation Substances 0.000 description 1
- 208000008443 pancreatic carcinoma Diseases 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000029058 respiratory gaseous exchange Effects 0.000 description 1
- 238000002560 therapeutic procedure Methods 0.000 description 1
- 210000001519 tissue Anatomy 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 238000011277 treatment modality Methods 0.000 description 1
- 210000004881 tumor cell Anatomy 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
- G06T7/344—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods involving models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/14—Transformations for image registration, e.g. adjusting or mapping for alignment of images
-
- 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]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention discloses a four-dimensional CT registration method and a four-dimensional CT registration device for a pancreatic region, wherein the four-dimensional CT registration method comprises the following steps: acquiring a fixed CT image and a mobile CT image of the abdomen; carrying out pancreas region segmentation on the fixed CT image by utilizing a pancreas segmentation model constructed based on a neural network to obtain a pancreas binary image; the pancreas binary image is used as a weight factor, the pancreas binary image, the fixed CT image and the mobile CT image are input into a registration model containing a plurality of registration sub-networks which are connected in sequence, a registered final deformation image is obtained through multiple times of registration calculation, the calculation speed of the registration process is high, and the pancreas binary image is used as the weight factor, so that the registration accuracy is improved.
Description
Technical Field
The invention belongs to the technical field of medical engineering, and particularly relates to a four-dimensional CT (computed tomography) registration method and device for a pancreatic region.
Background
Radiation therapy is one of the main technical means of tumor therapy at present. The radiation therapy method can eliminate and inhibit tumor cells in the target area by using high-energy X-rays and the like. In radiation therapy, respiratory-induced abdominal motion can lead to errors in treatment planning and delivery. Four-dimensional computed tomography (four-dimensional CT) with multiple scans over a respiratory cycle can achieve abdominal healthy tissue and tumor target tracking. Four-dimensional CT is increasingly used in radiation therapy to develop more accurate treatment plans. Accurate deformable registration on four-dimensional CT images facilitates treatment planning processes including motion tracking, target definition, risk organ preservation, and the like. Although viable registration has been widely studied for decades, current viable registration techniques still do not fully meet the increasingly demanding medical requirements.
Radiation therapy is one of the important treatment modalities for pancreatic cancer. During the respiration of the patient, the location of the pancreatic tumor can be severely affected by respiratory motion. Therefore, the four-dimensional CT technique is widely applied to radiotherapy of pancreatic tumors, and also increases difficulty in accurate pancreatic registration due to complicated and large-scale motion of the pancreas following the lungs.
Most of the conventional registration technology is realized based on an iterative algorithm, the speed is slow, and usually several minutes are needed for completing feasible variable registration of three-dimensional data once. For four-dimensional CT data, the time required to complete registration of all data is more unacceptable since the images typically have tens of phases. Therefore, it remains very challenging to develop accurate and fast registration for pancreatic regions.
Patent document CN111127527A discloses a method and device for realizing lung nodule adaptive matching based on CT image bone registration, including: preparing data, extracting point cloud data of lung and skeleton: registering two groups of skeleton three-dimensional point cloud data by adopting an FGR algorithm, and utilizing a transformation matrix obtained by registering the skeletons; evaluating registration errors of the two groups of lung point cloud data, and matching lung nodules by adopting a distance-based method based on the registration errors; the registration method is slow and the registration time is unacceptable.
Patent document CN113781593A discloses a method for generating a four-dimensional CT image, including: acquiring a reference three-dimensional CT image and a first three-dimensional CT image of a reference time phase corresponding to a reference respiratory state; carrying out rigid registration and deformation registration on the reference three-dimensional CT image and the first three-dimensional CT image of the reference time phase in sequence to obtain a translation parameter and a rotation parameter corresponding to the rigid registration and a deformation field corresponding to the deformation registration as image registration parameters; and transforming the first three-dimensional CT image of the residual time phase according to the image registration parameters and generating an intraoperative four-dimensional CT image. In the generation method, the adopted CT image registration mode is based on pixel-by-pixel calculation, so that the problem of low speed exists, and the registration time is difficult to accept.
Disclosure of Invention
In view of the above, the present invention provides a four-dimensional CT registration method and apparatus for pancreatic regions, which can achieve fast and accurate registration of four-dimensional CT.
To achieve the above object, an embodiment provides a four-dimensional CT registration method for a pancreatic region, including the following steps:
acquiring a fixed CT image and a mobile CT image of the abdomen;
carrying out pancreas region segmentation on the fixed CT image by utilizing a pancreas segmentation model constructed based on a neural network to obtain a pancreas binary image;
performing image registration based on a fixed CT image, a mobile CT image and a pancreas binary image by using a registration model comprising a plurality of registration sub-networks which are connected in sequence, wherein each registration sub-network comprises a calculation network connected with a deformation field and a spatial transformation function, and the image registration process of each registration sub-network is represented as follows:
wherein, I f Representing a fixed CT image, Bi is a binary image of the pancreas, phi n Representing the deformation field output by the nth registration sub-network, the sign deg. representing the interpolation calculation,
representing the deformation field phi n And moving CT image I m The deformation image obtained by interpolation calculation is realized through a spatial transformation function, and when n is equal to 1, I w As an input moving CT image, i.e. I m When n is>When the pressure of the mixture is 1, the pressure is lower,
representing the deformed image obtained by the (n-1) th registration sub-network, r n () represents the computing network that the nth registration subnetwork contains;
and taking the deformation image obtained by the last registration sub-network as a final deformation image output by the registration model.
In one embodiment, the pancreas segmentation model is constructed by:
acquiring an abdominal four-dimensional CT image, and marking a pancreas region on the abdominal four-dimensional CT image to obtain a pancreas region label;
and taking the abdominal four-dimensional CT image and the corresponding pancreatic region label as sample data, performing parameter optimization on the neural network in a supervised learning mode, and taking the neural network after parameter optimization as a pancreatic segmentation model.
In one embodiment, when the neural network is optimized, the neural network parameters are updated by taking the pancreas region prediction result output by the neural network and the Dice coefficient of the pancreas region label as loss functions.
In one embodiment, the registration model comprising a plurality of registration subnetworks connected in sequence needs to be optimized by model parameters before being applied, including:
taking a fixed CT image, a mobile CT image and a pancreas binary image obtained by acute segmentation of the fixed CT image by using a pancreas segmentation model as a training sample;
constructing a total loss function, including regularization loss of a deformation field constructed based on the deformation field output by the last registration subnetwork, similarity loss of a complete image constructed based on the fixed CT image and the final deformation image, and similarity loss of a pancreas region constructed based on the fixed CT image, the final deformation image, a pancreas binary image and a pancreas binary image through space transformation of the deformation field output by the last registration subnetwork;
and optimizing parameters of the registration sub-networks by using a total loss function and adopting an unsupervised learning mode, wherein the registration sub-networks after parameter optimization are used as registration models.
In one embodiment, the deformation field regularization loss is expressed as:
wherein the content of the first and second substances,
representing the deformation field phi output to the last registration subnetwork final The loss of regularization of (a) is,
representing gradient calculation, | · | | non-conducting phosphor 2 Representing the L2 norm.
In one embodiment, the complete image similarity loss and the pancreas region similarity loss are both expressed as root mean square errors:
wherein, I f A fixed CT image is represented as a fixed CT image,
represents the final deformation image, p represents pixels, Ω represents all pixels, Bi represents a binary image of the pancreas, Bi represents φ Representing the deformation field phi output by the last registration subnetwork final A spatially transformed binary image of the pancreas,
indicating a loss of similarity for the entire image,
indicating a loss of similarity of pancreatic regions;
the total loss function is expressed as:
wherein alpha, beta and gamma represent regularization coefficients of deformation field regularization loss, complete image similarity loss and pancreas region similarity loss, respectively,
the total loss function is represented.
In one embodiment, the registration model comprises a plurality of registration sub-networks with different network parameters and different network layer numbers, and each registration sub-network comprises a calculation network adopting a convolutional neural network.
To achieve the above object, the four-dimensional CT registration apparatus for pancreatic region provided by the embodiment includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the memory stores a pancreas segmentation model and a registration model optimized by parameters, and the processor implements the following steps when executing the computer program:
acquiring a fixed CT image and a mobile CT image of the abdomen;
carrying out pancreas region segmentation on the fixed CT image by utilizing a pancreas segmentation model constructed based on a neural network to obtain a pancreas binary image;
the image registration is carried out based on a fixed CT image, a mobile CT image and a pancreas binary image by utilizing a registration model containing a plurality of registration sub-networks which are connected in sequence, and the image registration comprises the following steps: the image registration process for each registration sub-network is represented as:
wherein, I f Representing a fixed CT image, Bi is a binary image of the pancreas, phi n Representing the deformation field output by the nth registration sub-network, the sign deg. representing the interpolation calculation,
representing the deformation field phi n And moving CT image I m When n is 1, I is the deformation image obtained by interpolation calculation w As input moving CT images, i.e. I m When n is>When the pressure of the mixture is 1, the pressure is lower,
representing the deformed image obtained by the (n-1) th registration subnetwork;
and taking the deformation image obtained by the last registration sub-network as a final deformation image output by the registration model.
To achieve the above object, an embodiment provides a four-dimensional CT registration apparatus for a pancreatic region, including:
the acquisition module is used for acquiring a fixed CT image and a mobile CT image of the abdomen;
the segmentation module is used for carrying out pancreas region segmentation on the fixed CT image by utilizing a pancreas segmentation model constructed based on a neural network so as to obtain a pancreas binary image;
the registration module is used for carrying out image registration based on a fixed CT image, a mobile CT image and a pancreas binary image by utilizing a registration model containing a plurality of registration sub-networks which are connected in sequence, and comprises the following steps: the image registration process for each registration sub-network is represented as:
wherein, I f Representing a fixed CT image, Bi is a binary image of the pancreas, phi n Representing the deformation field output by the nth registration sub-network, the sign deg. representing the interpolation calculation,
representing the deformation field phi n And moving CT image I m When n is 1, I is the deformation image obtained by interpolation calculation w As input moving CT images, i.e. I m When n is>When the pressure of the mixture is 1, the pressure is lower,
representing the deformed image obtained by the (n-1) th registration subnetwork;
and taking the deformation image obtained by the last registration sub-network as a final deformation image output by the registration model.
Compared with the prior art, the invention has the beneficial effects that at least:
on the basis that a pancreas segmentation model is used for carrying out pancreas region segmentation on a fixed CT image to obtain a pancreas binary image, the pancreas binary image is used as a weight factor, the fixed CT image and a mobile CT image are input into a registration model comprising a plurality of registration sub-networks which are connected in sequence, a registered final deformation image is obtained through multiple times of registration calculation, the calculation speed of the registration process is high, and the pancreas binary image is used as the weight factor, so that the registration accuracy is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
Fig. 1 is a flowchart of a four-dimensional CT registration method for a pancreatic region provided by an embodiment;
FIG. 2 is a two-value image of a pancreas obtained by pancreas segmentation on a fixed CT image according to an embodiment;
FIG. 3 is a schematic diagram of a registration model structure and registration calculation and training provided by an embodiment;
fig. 4 is a schematic structural diagram of a four-dimensional CT registration apparatus for a pancreatic region provided by an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the detailed description and specific examples, while indicating the scope of the invention, are intended for purposes of illustration only.
In order to achieve four-dimensional CT registration speed and accuracy of a pancreatic region, embodiments provide a four-dimensional CT registration method and apparatus for a pancreatic region.
Fig. 1 is a flowchart of a four-dimensional CT registration method for a pancreatic region according to an embodiment. As shown in fig. 1, the four-dimensional CT registration method for pancreatic regions provided by the embodiment includes the following steps:
step 1, acquiring a fixed CT image and a mobile CT image of the abdomen.
In an embodiment, the fixed CT image and the moving CT image of the abdomen may be from a clinical cone beam CT captured image, where the fixed CT image is a reference image, the moving CT image is an image to be registered, and both the fixed CT image and the moving CT image are three-dimensional grayscale images.
And 2, carrying out pancreas region segmentation on the fixed CT image by using a pancreas segmentation model constructed based on the neural network to obtain a pancreas binary image.
In an embodiment, the pancreas segmentation model is constructed by training a neural network, and the specific construction process includes:
and acquiring a standard abdominal four-dimensional CT image, and marking a pancreas region on the abdominal four-dimensional CT image to obtain a pancreas region label. During the mark, the clinician adopts manual mode to carry out the accuracy segmentation to the pancreas region to mark the pancreas region.
And taking the abdominal four-dimensional CT image and the corresponding pancreatic region label as sample data, performing parameter optimization on the neural network in a supervised learning mode, and taking the neural network after parameter optimization as a pancreatic segmentation model. Specifically, when the neural network is optimized, the neural network parameters are updated by taking the pancreas region prediction result output by the neural network and the Dice coefficient of the pancreas region label as loss functions.
When the pancreas is segmented, the fixed CT image is input into the pancreas segmentation model, the pancreas region of the fixed CT image is segmented through segmentation calculation, and the output segmentation result is used as a pancreas binary image. Illustratively, as shown in fig. 2, the fixed CT image shown on the left is subjected to pancreas region segmentation, resulting in a pancreas binary map shown on the right, where weight 0 in the pancreas binary map represents a non-pancreas region and weight 1 represents a pancreas region.
And 3, performing image registration based on the fixed CT image, the mobile CT image and the pancreas binary image by using a registration model comprising a plurality of registration sub-networks which are connected in sequence, and outputting a final deformation image.
In an embodiment, the registration model comprises a plurality of registration sub-networks connected in series, each registration sub-network registering the input image, each registration sub-network being an independent registration network having respective inputs and outputs. Each registration subnetwork comprises a computing network connected with the deformation field and a spatial transformation function, and realizes the registration process. Specifically, the image registration process for each registration sub-network is represented as:
wherein, I f Representing a fixed CT image, Bi is a binary image of the pancreas, phi n Representing the deformation field output by the nth registration sub-network, the sign deg. representing the interpolation calculation,
representing the deformation field phi n And moving CT image I m When n is 1, I is the deformation image obtained by interpolation calculation w As input moving CT images, i.e. I m When n is>When the pressure of the mixture is 1, the pressure is lower,
representing the deformed image obtained by the (n-1) th registration sub-network, r n (. cndot.) denotes the computing network that the nth registration subnetwork contains.
As shown in fig. 2, taking the registration model comprising three registration subnetworks as an example, the input of the first registration subnetwork is the input of the registration model, which includes the fixed CT image I f Moving CT image I m And a pancreas binary image Bi, and outputting the pancreas binary image Bi as a first deformation image after primary registration
The second registration sub-network is input as a stationary CT image I f First deformed image outputted from first registration sub-network
And a pancreas binary image Bi, and outputting the second deformation image
The third registration sub-network inputs a fixed CT image I f A second registration sub-network outputting a second deformation image
And a pancreas binary image Bi, a second deformation image is output
As the final deformation image of the registration model.
In an embodiment, a registration model including a plurality of registration subnetworks connected in sequence needs to be optimized by model parameters before being applied, including:
firstly, constructing a training sample, and taking a fixed CT image, a mobile CT image and a pancreas binary image obtained by acute segmentation of the fixed CT image by using a pancreas segmentation model as the training sample;
then, constructing a total loss function, including deformation field regularization loss constructed based on a deformation field output by the last registration subnetwork, complete image similarity loss constructed based on the fixed CT image and the final deformation image, and pancreas region similarity loss constructed based on the fixed CT image, the final deformation image, a pancreas binary image and a pancreas binary image through deformation field spatial transformation output by the last registration subnetwork;
wherein the deformation field regularization penalty is used to control the smoothness of the deformation field, avoid its excessive distortion, and produce an unrealistic deformation image, expressed as:
wherein the content of the first and second substances,
representing the deformation field phi output to the last registration subnetwork final The loss of regularization of (a) is,
represents gradient calculation, | Lu | | Liao 2 Represents the L2 norm;
the full image similarity loss is used to evaluate the similarity between the registered final deformed image and the fixed image, using the root mean square error, expressed as:
the pancreas region similarity loss is used to evaluate the similarity of the pancreas regions in the registered final deformed image and the fixed image, using root mean square error, expressed as:
wherein, I f Which represents a fixed CT image, is shown,
represents the final deformation image, p represents pixels, Ω represents all pixels, Bi represents the pancreas binary image φ Representing the deformation field phi output by the last registration subnetwork final A spatially transformed binary image of the pancreas,
indicating a loss of the similarity of the entire image,
indicating a loss of similarity of pancreatic regions;
thus, the total loss function is expressed as:
wherein alpha, beta and gamma respectively represent regularization coefficients of deformation field regularization loss, complete image similarity loss and pancreas region similarity loss,
the total loss function is represented.
And finally, optimizing parameters of a plurality of registration sub-networks by using a total loss function in an unsupervised learning mode, namely training without using a real deformation field as input, inputting a registration model into a fixed CT image, a mobile CT image and a pancreas binary image serving as a weight factor in the registration process, and outputting the output of the registration model which is the deformation image obtained by registration. And taking the registration sub-networks after parameter optimization as registration models.
When the trained registration model is used for image registration, the fixed CT image, the mobile CT image and the pancreas binary image are used as the input of the registration model, the registration processes of a plurality of registration sub-networks are carried out, the image registration for a plurality of times is realized, and the deformation image obtained by the last registration sub-network is used as the final deformation image output by the registration model.
The embodiment provides a four-dimensional CT registration device for a pancreatic region, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the memory stores a pancreas segmentation model and a registration model optimized by parameters, and the processor implements the following steps when executing the computer program:
step 1, acquiring a fixed CT image and a mobile CT image of an abdomen;
step 2, carrying out pancreas region segmentation on the fixed CT image by utilizing a pancreas segmentation model constructed based on a neural network to obtain a pancreas binary image;
and 3, performing image registration based on the fixed CT image, the mobile CT image and the pancreas binary image by using a registration model comprising a plurality of registration sub-networks which are connected in sequence, and outputting a final deformation image.
It should be noted that the pancreas segmentation model and the registration model optimized by the parameters stored in the four-dimensional CT registration apparatus are constructed by using the construction method included in the four-dimensional CT registration method provided in the above embodiment, and details are not repeated here. The specific process implemented in steps 1-3 of the four-dimensional CT registration apparatus is the same as the process of the four-dimensional CT registration method provided in the above embodiment, and is not described herein again.
In practical applications, the memory may be a volatile memory at the near end, such as RAM, a non-volatile memory, such as ROM, FLASH, a floppy disk, a mechanical hard disk, etc., or a remote storage cloud. The processor may be a Central Processing Unit (CPU), a microprocessor unit (MPU), a Digital Signal Processor (DSP), or a Field Programmable Gate Array (FPGA), i.e., a four-dimensional CT registration process for the pancreatic region may be implemented by these processors.
As shown in fig. 3, an embodiment provides a four-dimensional CT registration apparatus for a pancreatic region, including:
the acquisition module is used for acquiring a fixed CT image and a mobile CT image of the abdomen;
the segmentation module is used for carrying out pancreas region segmentation on the fixed CT image by utilizing a pancreas segmentation model constructed based on a neural network so as to obtain a pancreas binary image;
and the registration module is used for performing image registration on the basis of the fixed CT image, the mobile CT image and the pancreas binary image by using a registration model comprising a plurality of registration sub-networks which are connected in sequence, and outputting a final deformation image.
It should be noted that, when performing four-dimensional CT registration, the four-dimensional CT registration apparatus for a pancreatic region provided in the foregoing embodiment should be exemplified by the division of each functional module, and the function assignment may be completed by different functional modules as needed, that is, the internal structure of the terminal or the server is divided into different functional modules, so as to complete all or part of the functions described above. In addition, the four-dimensional CT registration apparatus for a pancreatic region provided in the above embodiment and the four-dimensional CT registration method for a pancreatic region belong to the same concept, and the specific implementation process thereof is described in detail in the four-dimensional CT registration method for a pancreatic region, and is not described herein again.
The above-mentioned embodiments are intended to illustrate the technical solutions and advantages of the present invention, and it should be understood that the above-mentioned embodiments are only the most preferred embodiments of the present invention, and are not intended to limit the present invention, and any modifications, additions, equivalents, etc. made within the scope of the principles of the present invention should be included in the scope of the present invention.
Claims (9)
1. A four-dimensional CT registration method for pancreatic regions, comprising the steps of:
acquiring a fixed CT image and a mobile CT image of the abdomen;
carrying out pancreas region segmentation on the fixed CT image by utilizing a pancreas segmentation model constructed based on a neural network to obtain a pancreas binary image;
performing image registration based on a fixed CT image, a mobile CT image and a pancreas binary image by using a registration model comprising a plurality of registration sub-networks which are connected in sequence, wherein each registration sub-network comprises a calculation network connected with a deformation field and a spatial transformation function, and the image registration process of each registration sub-network is represented as follows:
wherein, I f Representing a fixed CT image, Bi is a binary image of the pancreas, phi n Distortion field, sign representing the output of the nth registration subnetwork
It is indicated that the interpolation calculation is performed,
representing the deformation field phi n And moving CT image I m And when n is equal to 1, I is w As input moving CT images, i.e. I m When n is>When the pressure is 1, the pressure is higher,
representing the deformed image obtained by the (n-1) th registration sub-network, r n () represents the computing network that the nth registration subnetwork contains;
and taking the deformation image obtained by the last registration sub-network as a final deformation image output by the registration model.
2. The four-dimensional CT registration method for pancreatic regions according to claim 1, wherein the pancreatic segmentation model is constructed by:
acquiring an abdominal four-dimensional CT image, and marking a pancreas region on the abdominal four-dimensional CT image to obtain a pancreas region label;
and taking the abdominal four-dimensional CT image and the corresponding pancreatic region label as sample data, performing parameter optimization on the neural network in a supervised learning mode, and taking the neural network after parameter optimization as a pancreatic segmentation model.
3. The four-dimensional CT registration method for pancreatic regions according to claim 2, wherein in the parameter optimization of neural network, the neural network parameters are updated by using the Rice coefficients of pancreatic region labels and the pancreatic region prediction results outputted by neural network as loss functions.
4. The four-dimensional CT registration method for pancreatic regions according to claim 1, wherein the registration model containing a plurality of registration sub-networks connected in sequence needs to be optimized by model parameters before application, comprising:
taking a fixed CT image, a mobile CT image and a pancreas binary image obtained by acute segmentation of the fixed CT image by using a pancreas segmentation model as a training sample;
constructing a total loss function, including regularization loss of a deformation field constructed based on the deformation field output by the last registration subnetwork, similarity loss of a complete image constructed based on the fixed CT image and the final deformation image, and similarity loss of a pancreas region constructed based on the fixed CT image, the final deformation image, a pancreas binary image and a pancreas binary image through space transformation of the deformation field output by the last registration subnetwork;
and optimizing parameters of the registration sub-networks by using a total loss function in an unsupervised learning mode, wherein the registration sub-networks after parameter optimization serve as registration models.
5. The four-dimensional CT registration method for pancreatic regions of claim 1, wherein the deformation field regularization loss is expressed as:
wherein the content of the first and second substances,
representing the deformation field phi output to the last registration subnetwork final The loss of regularization of (a) is,
representing gradient calculation, | · | | non-conducting phosphor 2 Representing the L2 norm.
6. The four-dimensional CT registration method for pancreatic regions according to claim 1, wherein the complete image similarity loss and the pancreatic region similarity loss both use root mean square error, respectively expressed as:
wherein, I f A fixed CT image is represented as a fixed CT image,
represents the final deformation image, p represents pixels, Ω represents all pixels, Bi represents the pancreas binary image φ Representing the deformation field phi output by the last registration subnetwork final A spatially transformed binary image of the pancreas,
indicating a loss of similarity for the entire image,
indicating a loss of similarity of pancreatic regions;
the total loss function is expressed as:
wherein alpha, beta and gamma respectively represent regularization coefficients of deformation field regularization loss, complete image similarity loss and pancreas region similarity loss,
the total loss function is represented.
7. The four-dimensional CT registration method for the pancreatic region according to claim 1, wherein the registration model comprises a plurality of registration sub-networks with different network parameters and different network layer numbers, and each registration sub-network comprises a calculation network using a convolutional neural network.
8. A four-dimensional CT registration apparatus for pancreatic regions, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the memory stores a pancreas segmentation model and a registration model optimized by parameters, and the processor implements the following steps when executing the computer program:
acquiring a fixed CT image and a mobile CT image of the abdomen;
carrying out pancreas region segmentation on the fixed CT image by utilizing a pancreas segmentation model constructed based on a neural network to obtain a pancreas binary image;
the image registration is carried out based on a fixed CT image, a mobile CT image and a pancreas binary image by utilizing a registration model containing a plurality of registration sub-networks which are connected in sequence, and the image registration comprises the following steps: the image registration process for each registration sub-network is represented as:
wherein, I f Representing a fixed CT image, Bi is a binary image of the pancreas, phi n Distortion field, sign representing the output of the nth registration subnetwork
It is indicated that the interpolation calculation is performed,
representing the deformation field phi n And moving CT image I m When n is 1, I is the deformation image obtained by interpolation calculation w As input moving CT images, i.e. I m When n is>When the pressure of the mixture is 1, the pressure is lower,
representing the deformed image obtained by the (n-1) th registration subnetwork;
and taking the deformation image obtained by the last registration sub-network as a final deformation image output by the registration model.
9. A four-dimensional CT registration apparatus for a pancreatic region, comprising:
the acquisition module is used for acquiring a fixed CT image and a mobile CT image of the abdomen;
the segmentation module is used for carrying out pancreas region segmentation on the fixed CT image by utilizing a pancreas segmentation model constructed based on a neural network so as to obtain a pancreas binary image;
the registration module is used for carrying out image registration based on a fixed CT image, a mobile CT image and a pancreas binary image by utilizing a registration model containing a plurality of registration sub-networks which are connected in sequence, and comprises the following steps: the image registration process for each registration sub-network is represented as:
wherein, I f Represents a fixed CT image, Bi is a pancreas binary image, phi n A distorted field, sign, representing the output of the n-th registration sub-network
It is indicated that the interpolation calculation is performed,
represents the deformation field phi n And moving CT image I m When n is 1, I is the deformation image obtained by interpolation calculation w As input moving CT images, i.e. I m When n is>When the pressure is 1, the pressure is higher,
representing the deformed image obtained by the (n-1) th registration subnetwork;
and taking the deformation image obtained by the last registration sub-network as a final deformation image output by the registration model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210589627.XA CN115100252A (en) | 2022-05-26 | 2022-05-26 | Four-dimensional CT registration method and device for pancreatic region |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210589627.XA CN115100252A (en) | 2022-05-26 | 2022-05-26 | Four-dimensional CT registration method and device for pancreatic region |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115100252A true CN115100252A (en) | 2022-09-23 |
Family
ID=83288162
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210589627.XA Pending CN115100252A (en) | 2022-05-26 | 2022-05-26 | Four-dimensional CT registration method and device for pancreatic region |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115100252A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116523983A (en) * | 2023-06-26 | 2023-08-01 | 华南师范大学 | Pancreas CT image registration method integrating multipath characteristics and organ morphology guidance |
-
2022
- 2022-05-26 CN CN202210589627.XA patent/CN115100252A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116523983A (en) * | 2023-06-26 | 2023-08-01 | 华南师范大学 | Pancreas CT image registration method integrating multipath characteristics and organ morphology guidance |
| CN116523983B (en) * | 2023-06-26 | 2023-10-27 | 华南师范大学 | Pancreas CT image registration method integrating multipath characteristics and organ morphology guidance |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhou et al. | Deep learning of the sectional appearances of 3D CT images for anatomical structure segmentation based on an FCN voting method | |
| AU2019449137B2 (en) | sCT image generation using cyclegan with deformable layers | |
| CN108921851B (en) | Medical CT image segmentation method based on 3D countermeasure network | |
| CN112424835B (en) | System and method for image reconstruction | |
| JP2021131872A (en) | Atlas-based segmentation using deep learning | |
| CN110321920A (en) | Image classification method, device, computer readable storage medium and computer equipment | |
| JP2019507657A (en) | Fluence map generation method for radiation therapy | |
| CN112598649B (en) | 2D/3D spine CT non-rigid registration method based on generation of countermeasure network | |
| CN112785632B (en) | Cross-modal automatic registration method for DR and DRR images in image-guided radiotherapy based on EPID | |
| Kläser et al. | Deep boosted regression for MR to CT synthesis | |
| CN112614169A (en) | 2D/3D spine CT (computed tomography) level registration method based on deep learning network | |
| Fu et al. | iPhantom: a framework for automated creation of individualized computational phantoms and its application to CT organ dosimetry | |
| CN114792326A (en) | Surgical navigation point cloud segmentation and registration method based on structured light | |
| CN115100252A (en) | Four-dimensional CT registration method and device for pancreatic region | |
| CN111192268B (en) | Medical image segmentation model construction method and CBCT image bone segmentation method | |
| CN113989110A (en) | Lung image registration method and device, computer equipment and storage medium | |
| CN114511497A (en) | Imaging method and device applied to cone beam CT sparse sampling | |
| CN112164447A (en) | Image processing method, device, equipment and storage medium | |
| US11816832B2 (en) | Devices, systems, and methods for medical imaging | |
| Chen et al. | Dual-domain modulation for high-performance multi-geometry low-dose CT image reconstruction | |
| US20230031910A1 (en) | Apriori guidance network for multitask medical image synthesis | |
| CN112419348A (en) | Male pelvic cavity CT segmentation method based on multitask learning edge correction network | |
| Bai et al. | Progressive medical image annotation with convolutional neural network-based interactive segmentation method | |
| Alam et al. | Generalizable cone beam ct esophagus segmentation using in silico data augmentation | |
| CN113850710B (en) | Cross-modal medical image accurate conversion method |
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 |