CN107103618B - Lung 4D-CT multiphase image registration method based on regression prediction - Google Patents

Lung 4D-CT multiphase image registration method based on regression prediction Download PDF

Info

Publication number
CN107103618B
CN107103618B CN201710128508.3A CN201710128508A CN107103618B CN 107103618 B CN107103618 B CN 107103618B CN 201710128508 A CN201710128508 A CN 201710128508A CN 107103618 B CN107103618 B CN 107103618B
Authority
CN
China
Prior art keywords
image
regression
floating
lung
registration
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.)
Expired - Fee Related
Application number
CN201710128508.3A
Other languages
Chinese (zh)
Other versions
CN107103618A (en
Inventor
张煜
刘月亮
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.)
Southern Medical University
Original Assignee
Southern Medical University
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 Southern Medical University filed Critical Southern Medical University
Publication of CN107103618A publication Critical patent/CN107103618A/en
Application granted granted Critical
Publication of CN107103618B publication Critical patent/CN107103618B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G06T7/32—Determination of transform parameters for the alignment of images, i.e. image registration using correlation-based methods
    • 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
    • G06T2207/10076—4D tomography; Time-sequential 3D tomography
    • 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
    • G06T2207/30061—Lung

Abstract

The invention discloses a lung 4D-CT multiphase image registration method based on regression prediction, which comprises the steps of (1) reading lung 4D-CT data to obtain image groups composed of different phases; (2) selecting a certain phase image in the image group as a reference image, and selecting a certain phase image as a floating image; (3) registering the other phase images except the reference image and the floating image in the image group to the reference image to obtain corresponding deformation fields; (4) partitioning other phase images and corresponding deformation fields, and constructing an image appearance-deformation field regression model according to the blocks; (5) inputting the floating image blocks into a regression model, predicting an initial deformation field of the floating image, and obtaining an intermediate image; (6) registering the intermediate image with the reference image; (7) and (5) reselecting the next phase image in the image group, putting the floating image and the deformation field thereof into the training set, and repeating the steps (4) to (6). The method can improve the registration accuracy of the lung 4D-CT image.

Description

Lung 4D-CT multiphase image registration method based on regression prediction
Technical Field
The invention relates to the technical field of medical image processing, in particular to a lung 4D-CT multiphase image registration method based on regression prediction.
Background
The lung 4D-CT image has important guiding value for the diagnosis and treatment of lung cancer, and provides real image basis for lung tumor localization and lung cancer radiotherapy. Registration of lung 4D-CT has been widely used in lung image analysis, such as: lung segmentation, tracking of movement of lung organ tissue, monitoring of disease, and the like.
However, there are local large deformations due to lung respiratory motion, and interference caused by lung texture and heart beat in the lung images; together with the assumption that the gray scale does not change, the theory does not hold completely for lung images. Therefore, the registration of the lung images is easy to have local misregistration and fall into local extrema.
Therefore, aiming at the existing insufficient problem of lung 4D-CT image registration, a lung 4D-CT multiphase image registration method based on regression prediction is provided.
Disclosure of Invention
The invention aims to provide a lung 4D-CT multiphase image registration method based on regression prediction, which utilizes image information of different phases corresponding to a floating image to predict a deformation field and can improve the registration accuracy of the lung 4D-CT image.
The aim of the invention is achieved by the following technical measures: a lung 4D-CT multiphase image registration method based on regression prediction comprises the following steps:
(1) reading lung 4D-CT data to obtain image groups composed of different phases;
(2) selecting a certain phase image in the image group as a reference image, and selecting a certain phase image as a floating image;
(3) registering the reference image removed from the image group and other phase images of the floating image to the reference image to obtain deformation fields corresponding to the other phase images of the reference image removed from the image group and the floating image;
(4) partitioning other phase images and corresponding deformation fields, and constructing an image appearance-deformation field regression model according to the blocks;
(5) inputting the floating image blocks selected in the step (2) into the regression model constructed in the step (4), predicting an initial deformation field of the floating image, and obtaining an intermediate image;
(6) registering the intermediate image obtained in step (5) with the reference image selected in step (2);
(7) and (3) reselecting the next phase image in the image group, putting the floating image and the deformation field thereof in the step (2) into a training set, repeating the steps (4) to (6), completing the predictive registration, and circularly registering until all the phase images in the image group are registered.
In the present invention, the registration in step (3) specifically adopts an Active Demons registration algorithm.
In the invention, the step (4) is to construct a training set by using other phase images and corresponding deformation field blocks, and then establish an image appearance-deformation field regression model by using a multi-dimensional support vector regression machine, and the specific process is as follows:
(4.1) the block strategy is a scanning type overlapping method from left to right and from top to bottom, and the block size is 32 pixels multiplied by 32 pixels;
(4.2) the pixel gray scale of the image block after being partitioned is used as a training input X of the regression model, and the deformation field of the image block is used as a training output Y; solving the image appearance-deformation field regression model by using the known training sets X and Y:
formula (1);
in the formula, R represents a regression relationship;
(4.3) solving the regression model by adopting a multi-dimensional support vector regression machine, wherein the multi-dimensional support vector regression machine can model a high-dimensional nonlinear mapping function and can simultaneously and independently predict each output dimension so as to fully utilize the spatial correlation, and the multi-dimensional support vector regression machine aims at learning the following regression functions:
formula (2);
wherein: phi (X): Rm→RhIs a non-linear mappingMapping from m dimensions to a higher dimension h space. From the known training data samples D { (X)i,Yi) I ═ 1,2,. n }, the main optimization parameter W ═ W ] of the regression function is solved1，w2,...wl]TAnd b ═ b1，b2,...bl]T，X∈RmIs the gray scale input of each pixel point of the image, Y belongs to RlOutputting the corresponding deformation of each pixel point of the image, wherein n is the number of training samples, and m, l and h all represent spatial dimensions;
(4.4) the solution of W and b is achieved by the following optimization problem:
Figure BDA0001239235220000021
Figure BDA0001239235220000022
in the formula ui＝||Yi-WΦ(Xi) -b | | is the prediction error, l (u) is the loss function, C is a constant, controlling the compromise between the penalty for out-of-error sample and the flatness of the function, is the prediction accuracy range;
the optimal solution of W is a linear combination of training samples in a high-dimensional feature space, namely:
Figure BDA0001239235220000031
the optimal W and beta are linear combination coefficients by an iterative variable weight least square method.
In the invention, the step (5) is to predict the block-shaped deformation field in blocks according to the regression model obtained in the step (4), and finally splice the block-shaped deformation field into the whole initial deformation field, and the specific process is as follows:
(5.1) blocking the floating image, inputting block appearance X into a regression model:
Figure BDA0001239235220000032
thus, the image block-shaped variable field output Y, K (X) can be obtainediX) represents a kernel function, superscript T represents transposition;
and (5.2) after the blocks are divided, sequentially predicting block deformation fields according to the formula (6), and finally splicing the block deformation fields into initial deformation fields corresponding to the floating images, thereby obtaining intermediate images.
In the invention, the step (6) is to refine and register the intermediate image and the reference image by using an Active Demons registration algorithm according to the intermediate image obtained in the step (5).
The same reference numerals are used throughout the specification to denote the same meanings.
The lung 4D-CT image registration method based on regression prediction utilizes the similarity of respiratory motion of lung 4D-CT multiphase images and uses image information of each phase corresponding to a floating image to predict a deformation field, thereby reducing the obvious deformation between the floating image and a reference image and improving the registration accuracy.
Drawings
The invention is described in detail with reference to the accompanying drawings and detailed description.
FIG. 1 is a flowchart of a regression prediction based lung 4D-CT multiphase image registration method of the present invention.
Fig. 2A to 2F are coronal reference images, coronal floating images, and the results of registration and the respective corresponding difference images using different methods according to the present invention, wherein:
FIG. 2A is a reference image;
FIG. 2B is a floating image;
FIG. 2C is a diagram of Active Demons registration results;
FIG. 2D is an Active Demons registration difference map;
fig. 2E is a graph of the registration result of the present invention.
Fig. 2F is a registration difference map of the present invention.
Fig. 3A to 3F are sagittal reference images, sagittal floating images, and the results of registration and the respective corresponding difference images using different methods according to the present invention, wherein:
FIG. 3A is a reference image;
FIG. 3B is a floating image;
FIG. 3C is a diagram of Active Demons registration results;
FIG. 3D is an Active Demons registration difference map;
FIG. 3E is a graph of the registration results of the present invention;
fig. 3F is a registration difference map of the present invention.
Detailed Description
FIG. 1 shows a flow chart of the method of the present invention, and the processing of the present invention is described in detail in connection with a publicly available set of lung 4D-CT data sets. The invention relates to a lung 4D-CT multiphase image registration method based on regression prediction, which comprises the following steps of:
(1) reading lung 4D-CT data to obtain an image group consisting of 10 different phase images;
(2) selecting one phase image as a reference image and the other phase image as a floating image in the image group;
(3) registering the reference image and other phase images of the floating image in the image group to the reference image by adopting an Active Demons registration algorithm to obtain deformation fields corresponding to the other phase images of the reference image and the floating image in the image group;
(4) establishing an image appearance-deformation field regression model by using other phase images and corresponding deformation field blocks as a training set and utilizing a multi-dimensional support vector regression mechanism;
step (4) is that other phase images and corresponding deformation field blocks are firstly constructed to form a training set, and then an image appearance-deformation regression model is constructed by utilizing a Multi-dimensional Support Vector regression (MSVR), wherein the specific process is as follows:
(4.1) the blocking strategy is a left-to-right, top-to-bottom scanning overlap method. The block size is 32 pixels × 32 pixels;
(4.2) the pixel gray scale of the image block after being partitioned is used as a training input X of the regression model, and the deformation field of the image block is used as a training output Y; solving the image appearance-deformation field regression model by using the known training sets X and Y:
formula (1);
in the formula, R represents a regression relationship;
(4.3) solving the regression model by adopting a multi-dimensional support vector regression machine, wherein the MSVR can model a high-dimensional nonlinear mapping function and can predict each output dimension simultaneously and independently so as to make full use of spatial correlation, and the MSVR aims at learning the following regression functions:
formula (2);
wherein: phi (X): Rm→RhIs a non-linear mapping from m dimensions to a higher dimension h space. From the known training data samples D { (X)i,Yi) I ═ 1,2,. n }, the main optimization parameter W ═ W ] of the regression function is solved1，w2,...wl]TAnd b ═ b1，b2,...bl]T，X∈RmIs the gray scale input of each pixel point of the image, Y belongs to RlAnd outputting deformation corresponding to each pixel point of the image, wherein n is the number of training samples, and m, l and h all represent spatial dimensions.
(4.4) the solution of W and b is achieved by the following optimization problem:
Figure BDA0001239235220000051
Figure BDA0001239235220000052
in the formula ui＝||Yi-WΦ(Xi) -b | | is the prediction error, l (u) is the loss function, C is a constant, controlling the compromise between the penalty for out of error limit samples and the flatness of the function, is the prediction accuracy range.
The optimal solution of W is a linear combination of training samples in a high-dimensional feature space, namely:
Figure BDA0001239235220000053
the optimal W and beta are linear combination coefficients by an iterative variable weight least square method.
(5) Inputting the floating image blocks selected in the step (2) into the regression model constructed in the step (4), predicting an initial deformation field of the floating image, and obtaining an intermediate image;
and (5) predicting the block-shaped deformation field in a blocking mode according to the regression model obtained in the step (4), and finally splicing the block-shaped deformation field into the whole initial deformation field. The specific process is as follows:
(5.1) blocking the floating image, and inputting the block appearance into a regression model:
Figure BDA0001239235220000054
thus, the image block-shaped variable field output Y, K (X) can be obtainediX) denotes the kernel function and the superscript T denotes the transpose.
And (5.2) after the blocks are divided, sequentially predicting block deformation fields according to the formula (6), and finally splicing the block deformation fields into initial deformation fields corresponding to the floating images, thereby obtaining intermediate images.
(6) Registering the intermediate image obtained in step (5) with the reference image selected in step (2);
and (6) refining and registering the intermediate image and the reference image by using an Active Demons registration algorithm according to the intermediate image obtained in the step (5).
(7) And (3) reselecting a phase image in the next image group, putting the floating image and the deformation field thereof in the step (2) into a training set, repeating the steps (4) to (6), completing the predictive registration, and circularly registering until all the phase images in the image group are registered.
Fig. 2A to 2F show a coronal reference image, a coronal floating image, and the results of different registration methods and their respective difference images, respectively, according to the present invention. Fig. 3A to 3F are sagittal plane reference images, sagittal plane floating images, and the results of different registration methods and the respective corresponding differential images of the present invention.
Besides the visual effect display, the invention also provides a quantification result for verifying the effectiveness of the method. The method adopts the sum of squared errors (SSD) of the images to evaluate the registration result of the floating images with the phase 0 as a reference image and other phases in turn.
The sum of the squared errors of the images is defined as:
Figure BDA0001239235220000061
where N is the total number of image pixels, FiFor reference image gray scale, TiIs the registered image gray scale. The smaller the SSD value is, the higher the image similarity is, and the better the registration effect is.
The average SSD registered to the sagittal plane of the phase 0 crown using the ActiveDemons algorithm and the method of the present invention, respectively, is calculated using equation (7), as shown in table 1 below. Therefore, the mean square error sum of the method is obviously lower than that of an Active Demons algorithm, and the registration method is more accurate.
Table 1: active Demons algorithm and method averaging SSD by using phase 0 as reference image and other phases as floating images
Figure BDA0001239235220000062
Figure BDA0001239235220000071
As can be seen from fig. 2D and fig. 2F, the difference images of different registration methods of the coronal plane have better registration effect for the lung image with larger deformation, and are obviously better than the Active Demons algorithm. Fig. 3D and 3F are differential images of different registration methods in the sagittal plane, respectively, and the same conclusion can be obtained. Table 1 also reflects the accuracy of the inventive method for lung registration.
It should be noted that the embodiments of the present invention are not limited thereto, and may be modified according to actual needs to adapt to different actual needs. Therefore, many other modifications, substitutions and alterations of the present disclosure will be apparent to those skilled in the art based on the foregoing basic technical ideas and their conventional practices, and are intended to be within the scope of the present disclosure.

Claims (5)

1. A lung 4D-CT multiphase image registration method based on regression prediction comprises the following steps:
(1) reading lung 4D-CT data to obtain image groups composed of different phases;
(2) selecting a certain phase image in the image group as a reference image, and selecting a certain phase image as a floating image;
(3) registering the reference image removed from the image group and other phase images of the floating image to the reference image to obtain deformation fields corresponding to the other phase images of the reference image removed from the image group and the floating image;
(4) constructing a training set by using other phase images and corresponding deformation field blocks thereof, and establishing an image appearance-deformation field regression model by using a multi-dimensional support vector regression machine;
(5) inputting the floating image blocks selected in the step (2) into the regression model constructed in the step (4), predicting an initial deformation field of the floating image, and obtaining an intermediate image;
(6) registering the intermediate image obtained in step (5) with the reference image selected in step (2);
(7) and (3) selecting the next phase image in the image group as a new floating image, putting the floating image in the step (2) and the deformation field thereof into a training set, repeating the steps (4) to (6), and completing the predictive registration and the cyclic registration until all the phase images in the image group are registered.
2. The regression prediction based lung 4D-CT multiphase image registration method according to claim 1, wherein: the registration in the step (3) specifically adopts an Active Demons registration algorithm.
3. The regression prediction based lung 4D-CT multiphase image registration method according to claim 1, wherein the step (4) is implemented as follows:
(4.1) the blocking strategy is a scanning type overlapping method from left to right and from top to bottom, and the size of the block is 32 pixels x32 pixels;
(4.2) the pixel gray scale of the image block after being partitioned is used as a training input X of the regression model, and the deformation field of the image block is used as a training output Y; solving the image appearance-deformation field regression model by using the known training sets X and Y:
y ═ r (x) … … formula (1);
in the formula, R represents a regression relationship;
(4.3) solving the regression model by adopting a multi-dimensional support vector regression machine, wherein the multi-dimensional support vector regression machine can model a high-dimensional nonlinear mapping function and can simultaneously and independently predict each output dimension so as to fully utilize the spatial correlation, and the multi-dimensional support vector regression machine aims at learning the following regression functions:
y ═ W Φ (X) + b … … formula (2);
wherein: φ (X): rm→RhIs a non-linear mapping from m dimensions to a higher dimension h space from known training data samples D { (X)i，Yi) I 1,2, n, solving the main optimization parameter W ═ W [ W ] of the regression function1，w2，...wl]TAnd b ═ b1，b2，...bl]T，X∈RmIs the gray scale input of each pixel point of the image, Y belongs to RlOutputting the corresponding deformation of each pixel point of the image, wherein n is the number of training samples, and m, l and h all represent spatial dimensions;
(4.4) the solution of w and b is achieved by the following optimization problem:
Figure FDA0002576606220000011
Figure FDA0002576606220000021
in the formula ui＝||Yi-Wφ(Xi) -b | | is the prediction error, l (u) is the loss function, C is a constant, controlling the compromise between the penalty for out-of-error sample and the flatness of the function, is the prediction accuracy range;
the optimal solution of W is a linear combination of training samples in a high-dimensional feature space, namely:
Figure FDA0002576606220000022
the optimal W and beta are linear combination coefficients by an iterative variable weight least square method.
4. The regression prediction based lung 4D-CT multiphase image registration method according to claim 3, wherein: and (5) predicting the block-shaped deformation field in a blocking mode according to the regression model obtained in the step (4), and finally splicing the block-shaped deformation field into the whole initial deformation field, wherein the specific process is as follows:
(5.1) blocking the floating image, inputting block appearance X into a regression model:
Figure FDA0002576606220000023
thus, the image block-shaped variable field output Y, K (X) can be obtainediX) represents a kernel function, superscript T represents transposition;
and (5.2) after the blocks are divided, sequentially predicting block deformation fields according to the formula (6), and finally splicing the block deformation fields into initial deformation fields corresponding to the floating images, thereby obtaining intermediate images.
5. The regression prediction based lung 4D-CT multiphase image registration method according to claim 1, wherein: and (6) refining and registering the intermediate image and the reference image by using an Active Demons registration algorithm according to the intermediate image obtained in the step (5).
CN201710128508.3A 2017-02-20 2017-03-06 Lung 4D-CT multiphase image registration method based on regression prediction Expired - Fee Related CN107103618B (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710088559 2017-02-20
CN2017100885598 2017-02-20

Publications (2)

Publication Number Publication Date
CN107103618A CN107103618A (en) 2017-08-29
CN107103618B true CN107103618B (en) 2020-10-27

Family

ID=59675188

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710128508.3A Expired - Fee Related CN107103618B (en) 2017-02-20 2017-03-06 Lung 4D-CT multiphase image registration method based on regression prediction

Country Status (1)

Country Link
CN (1) CN107103618B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11880988B2 (en) 2020-06-11 2024-01-23 GE Precision Healthcare LLC Image registration method and model training method thereof

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111489381B (en) * 2019-01-28 2023-11-10 临沂大学 Infant brain nuclear magnetic resonance image group registration method and device
CN111260700B (en) * 2020-01-09 2023-05-30 复旦大学 Full-automatic registration and segmentation method for multi-parameter magnetic resonance image
CN116977387B (en) * 2023-09-22 2023-12-15 安徽大学 Deformable medical image registration method based on deformation field fusion

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103886568A (en) * 2014-03-18 2014-06-25 南方医科大学 Lung 4D-CT image super-resolution reconstruction method based on registration
CN104361567A (en) * 2014-09-18 2015-02-18 南方医科大学 Lung 4D-CT image inhaling process middle phase image reconstruction method based on registration
CN104361568A (en) * 2014-09-18 2015-02-18 南方医科大学 Lung 4D-CT image exhaling process middle phase image reconstruction method based on registration
CN104599268A (en) * 2015-01-06 2015-05-06 广州医科大学附属肿瘤医院 Local area accurate deformation registration algorithm combining point registration
CN105303547A (en) * 2014-07-11 2016-02-03 东北大学 Multiphase CT image registration method based on grid matching Demons algorithm
CN105913442A (en) * 2016-04-29 2016-08-31 上海联影医疗科技有限公司 Method for automatically matching pulmonary nodules

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103886568A (en) * 2014-03-18 2014-06-25 南方医科大学 Lung 4D-CT image super-resolution reconstruction method based on registration
CN105303547A (en) * 2014-07-11 2016-02-03 东北大学 Multiphase CT image registration method based on grid matching Demons algorithm
CN104361567A (en) * 2014-09-18 2015-02-18 南方医科大学 Lung 4D-CT image inhaling process middle phase image reconstruction method based on registration
CN104361568A (en) * 2014-09-18 2015-02-18 南方医科大学 Lung 4D-CT image exhaling process middle phase image reconstruction method based on registration
CN104599268A (en) * 2015-01-06 2015-05-06 广州医科大学附属肿瘤医院 Local area accurate deformation registration algorithm combining point registration
CN105913442A (en) * 2016-04-29 2016-08-31 上海联影医疗科技有限公司 Method for automatically matching pulmonary nodules

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
A General Fast Registration Framework by Learning Deformation–Appearance Correlation;Minjeong Kim et al;《IEEE Transactions on Image Processing》;20120430;第21卷(第4期);第1823页至第1829页 *
Adaptive patch-based POCS approach for super resolution reconstruction of 4D-CT lung data;Tingting Wang et al;《PHYSICS IN MEDICINE AND BIOLOGY》;20150807;第60卷(第15期);全文 *
Improved image registration by sparse patch-based deformation estimation;Minjeong Kim et al;《NeuroImage》;20150115;第105卷;全文 *
Regression Segmentation for M3 Spinal Images;Zhijie Wang et al;《IEEE Transactions on Medical Imaging》;20150831;第34卷(第8期);第1640页至第1644页 *
基于四维CT图像形变配准技术的研究及其在放射性肝损伤预测中的应用;王惠;《中国优秀硕士学位论文全文数据库 信息科技辑》;20131115;第22页至第34页 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11880988B2 (en) 2020-06-11 2024-01-23 GE Precision Healthcare LLC Image registration method and model training method thereof

Also Published As

Publication number Publication date
CN107103618A (en) 2017-08-29

Similar Documents

Publication Publication Date Title
CN107103618B (en) Lung 4D-CT multiphase image registration method based on regression prediction
Bratt et al. Machine learning derived segmentation of phase velocity encoded cardiovascular magnetic resonance for fully automated aortic flow quantification
US9561004B2 (en) Automated 3-D orthopedic assessments
CN110021037A (en) A kind of image non-rigid registration method and system based on generation confrontation network
CN109377520B (en) Heart image registration system and method based on semi-supervised circulation GAN
Nambakhsh et al. Left ventricle segmentation in MRI via convex relaxed distribution matching
CN104021547A (en) Three dimensional matching method for lung CT
CN112634265B (en) Method and system for constructing and segmenting fully-automatic pancreas segmentation model based on DNN (deep neural network)
Liu et al. Coordinate translator for learning deformable medical image registration
Li et al. Mdreg‐net: Multi‐resolution diffeomorphic image registration using fully convolutional networks with deep self‐supervision
Wang et al. Sam-med3d
Bartoli Maximizing the predictivity of smooth deformable image warps through cross-validation
Ristea et al. CyTran: a cycle-consistent transformer with multi-level consistency for non-contrast to contrast CT translation
Zhang et al. A diffeomorphic unsupervised method for deformable soft tissue image registration
Ma et al. Unsupervised deformable image registration network for 3D medical images
Gass et al. Consistency-based rectification of nonrigid registrations
CN110517300A (en) Elastic image registration algorithm based on partial structurtes operator
Fourcade et al. Deformable image registration with deep network priors: a study on longitudinal PET images
He et al. Unsupervised end-to-end groupwise registration framework without generating templates
Ng et al. Surface-based mapping of gene expression and probabilistic expression maps in the mouse cortex
CN105631897A (en) Movie nuclear magnetic resonance image sequence motion estimation method based on monogenic signal characteristic distance and cross-correlation transform optical flow algorithm
Kher Implementation of image registration for satellite images using mutual information and particle swarm optimization techniques
Ma et al. SEN-FCB: an unsupervised twinning neural network for image registration
CN112435284A (en) Image registration method, electronic device and storage medium
Zhang et al. Intensity based image registration by minimizing exponential function weighted residual complexity

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20201027

Termination date: 20210306