CN108537723B - Three-dimensional nonlinear registration method and system for massive brain image data sets - Google Patents
Three-dimensional nonlinear registration method and system for massive brain image data sets Download PDFInfo
- Publication number
- CN108537723B CN108537723B CN201810307540.2A CN201810307540A CN108537723B CN 108537723 B CN108537723 B CN 108537723B CN 201810307540 A CN201810307540 A CN 201810307540A CN 108537723 B CN108537723 B CN 108537723B
- Authority
- CN
- China
- Prior art keywords
- resolution
- registration
- image data
- low
- transformation
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 28
- 210000004556 brain Anatomy 0.000 title claims abstract description 21
- 230000009466 transformation Effects 0.000 claims abstract description 103
- 239000011159 matrix material Substances 0.000 claims abstract description 83
- 230000000903 blocking effect Effects 0.000 claims abstract description 27
- 238000006073 displacement reaction Methods 0.000 claims abstract description 27
- 238000005070 sampling Methods 0.000 claims abstract description 14
- 239000013598 vector Substances 0.000 claims description 24
- 238000013507 mapping Methods 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 5
- 238000010586 diagram Methods 0.000 description 3
- 230000001131 transforming effect Effects 0.000 description 3
- 238000000605 extraction Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000000638 solvent extraction Methods 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- 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
- G06T3/147—Transformations for image registration, e.g. adjusting or mapping for alignment of images using affine transformations
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a three-dimensional nonlinear registration method and a system of a massive brain image data set, wherein the three-dimensional nonlinear registration method of the massive brain image data set comprises the following steps: s0. down-sampling; s1, a low-resolution image registration step, which comprises the following steps: s11, a linear registration step; s12, a nonlinear registration step; s2, a high resolution image fast transformation step, which comprises the following steps: s21, calculating a transformation matrix of high-resolution registration; s22, calculating a space range after high-resolution registration; s23, a blocking step; s24, high-resolution nonlinear transformation. The invention utilizes the low-resolution image data registration information to transform the massive high-resolution three-dimensional image data set, thereby realizing the nonlinear registration of the massive high-resolution three-dimensional image data set. The method is suitable for all linear registration based on the transformation matrix and all nonlinear registration based on the displacement field, and can rapidly carry out nonlinear registration on the TB-level data set.
Description
Technical Field
The invention relates to an image processing technology, in particular to a three-dimensional nonlinear registration method of a massive brain image data set.
Background
As an important research direction in the field of image processing, image registration technology is widely applied in the fields of remote sensing, automatic navigation, pattern recognition, computer vision, biomedicine, and the like. With the advent of the big data era, massive three-dimensional image data sets are continuously increased and can reach TB level. How to perform non-linear registration on these massive three-dimensional data sets is a very big challenge.
In the chinese invention patent specification CN104361590, a high-resolution remote sensing image registration method with control points adaptively distributed is proposed, which is based on a self-adaptive control point extraction method of a blocking strategy, realizes control point extraction in a multi-scale J-image, and further performs multi-scale matching on the control points by using normalized mutual information measure NMI. However, the data processed by the method is two-dimensional remote sensing data and is suitable for specific application fields. Moreover, the high resolution described by the method is only the spatial resolution of 10m, the size is 512 × 512 pixels, the data size is only in the MB level, and the non-linear registration cannot be performed on TB level three-dimensional image data.
Tools widely used in image Registration studies also enable non-linear Registration of three-dimensional datasets, such as the Insight Segmentation and Registration Toolkit (ITK) and Advanced Normalization Tools (ANTs). Although these tools provide rich three-dimensional non-linear registration methods, they consider three-dimensional images as a whole in the registration process and once lead into a computer for operation, which sharply increases the consumption of memory. For three-dimensional image data of dozens of MB, some tools consume even hundreds of GB of memory, and the non-linear registration of three-dimensional image data of TB level is more difficult.
By combining the patent technology and tools, at present, a plurality of methods exist for the nonlinear registration of three-dimensional images, and the problems in various fields can be solved. However, with the coming of big data era, massive three-dimensional image data sets are continuously increased, and a non-linear registration method capable of processing TB-level three-dimensional image data does not exist.
Disclosure of Invention
The invention aims to solve the technical problems and provides a three-dimensional nonlinear registration method and a three-dimensional nonlinear registration system for massive brain image data sets, which are used for rapidly transforming the massive high-resolution three-dimensional image data sets and realizing the nonlinear registration of the massive high-resolution three-dimensional image data sets.
The object of the present invention is achieved by the following means.
A three-dimensional non-linear registration method for a massive brain image data set comprises the following steps:
s0., a down-sampling step, namely down-sampling preset high-resolution reference image data and high-resolution image data to be registered to obtain low-resolution reference image data and low-resolution image data to be registered;
s1, a low-resolution image registration step, which comprises the following steps:
s11, a linear registration step, namely performing linear registration on the low-resolution reference image data and the low-resolution image data to be registered to obtain a transformation matrix of the low-resolution linear registration, and calculating an inverse matrix of the low-resolution linear registration and a result after the low-resolution linear registration;
s12, a nonlinear registration step, namely performing nonlinear registration on the result after the low-resolution linear registration and the low-resolution reference image data to obtain a displacement field of the low-resolution nonlinear registration and a result after the low-resolution nonlinear registration;
s2, a high resolution image fast transformation step, which comprises the following steps:
s21, calculating a transformation matrix of high-resolution registration, namely calculating the transformation matrix of high-resolution registration and an inverse matrix of inverse matrix high-resolution registration based on the transformation matrix of low-resolution linear registration;
s22, calculating a space range after high-resolution registration, namely calculating the space range after high-resolution registration by using the inverse matrix of high-resolution registration obtained in the step S21;
s23, a blocking step, namely blocking the high-resolution space range obtained in the step S22 after registration;
and S24, a high-resolution nonlinear transformation step, namely performing parallel computation on each block obtained in the step S23, computing the corresponding position of each pixel point in the original high-resolution image data to be registered, and finally obtaining a high-resolution image nonlinear transformation result.
Further, the linear registration described in step S11 describes the transformation mapping as a transformation matrix.
Further, the non-linear registration described in step S12 describes the transformation mapping as a displacement field of pixel offset vectors.
Further, the blocking in step S23 is performed based on the computing resources, and the transformation of the pixel points in the blocking is calculated in parallel, where the computing resources can support n threads at maximum to perform parallel processing, and the blocking is performed with a length n.
Further, the high resolution nonlinear transformation in step S24 includes the following sub-steps:
s241, estimating the range of the original high-resolution image data to be registered before block corresponding transformation;
and S242, traversing each pixel point in the block, calculating an offset vector corresponding to the pixel point through interpolation based on a displacement field of low-resolution nonlinear registration, calculating a corresponding position of the pixel point in original high-resolution image data to be registered based on the offset vector and an inverse matrix of high-resolution registration, and taking a gray value of the pixel point as a gray value of the pixel point.
The invention also discloses a three-dimensional nonlinear registration system of the massive brain image data set, which comprises the following modules:
the down-sampling module is used for down-sampling preset high-resolution reference image data and high-resolution image data to be registered to obtain low-resolution reference image data and low-resolution image data to be registered;
a low resolution image registration module, comprising:
the linear registration unit is used for carrying out linear registration on the low-resolution reference image data and the low-resolution image data to be registered to obtain a transformation matrix of the low-resolution linear registration, and calculating an inverse matrix of the low-resolution linear registration and a result after the low-resolution linear registration;
the nonlinear registration unit is used for carrying out nonlinear registration on the low-resolution linear registration result and the low-resolution reference image data to obtain a displacement field of the low-resolution nonlinear registration and a low-resolution nonlinear registration result;
the high resolution image fast transformation module comprises:
the high-resolution registration transformation matrix calculation unit is used for calculating a high-resolution registration transformation matrix and an inverse matrix of the inverse matrix high-resolution registration transformation matrix based on the low-resolution linear registration transformation matrix;
the high-resolution post-registration spatial range calculating unit calculates a high-resolution post-registration spatial range by using the inverse matrix of the high-resolution registration obtained by the high-resolution registration transformation matrix calculating unit;
the blocking unit is used for blocking the high-resolution post-registration spatial range obtained by the high-resolution post-registration spatial range calculating unit;
and the high-resolution nonlinear transformation unit is used for performing parallel calculation on each block obtained by the block dividing unit, calculating the corresponding position of each pixel point in the original high-resolution image data to be registered, and finally obtaining a high-resolution image nonlinear transformation result.
Further, the linear registration in the linear registration unit describes the transformation mapping as a transformation matrix.
Further, the non-linear registration described in the non-linear registration unit describes the transformation mapping as a displacement field of pixel offset vectors.
Furthermore, the blocks in the block partitioning unit are partitioned based on computing resources, the conversion of pixel points in the blocks is calculated in parallel, the computing resources can support n threads to be parallel at most, and the blocks are partitioned according to the length n.
Further, the high resolution nonlinear transformation in the high resolution nonlinear transformation unit includes the following units:
the estimation unit estimates the range of the original high-resolution image data to be registered before the block corresponding transformation;
and the transformation unit is used for traversing each pixel point in the blocks, calculating an offset vector corresponding to the pixel point through interpolation based on the displacement field of low-resolution nonlinear registration, calculating the corresponding position of the pixel point in the original high-resolution image data to be registered based on the offset vector and the inverse matrix of high-resolution registration, and taking the gray value of the pixel point as the gray value of the pixel point.
The invention has the advantages that: and transforming the massive high-resolution three-dimensional image data set by using the low-resolution image data registration information to realize the nonlinear registration of the massive high-resolution three-dimensional image data set. The method is suitable for all linear registration based on the transformation matrix and all nonlinear registration based on the displacement field, and can rapidly carry out nonlinear registration on the TB-level data set.
Drawings
FIG. 1 is a schematic flow chart of a method for three-dimensional non-linear registration of a large volume of brain image data sets in accordance with a preferred embodiment of the present invention;
FIG. 2 is a schematic diagram of fast transformation of high resolution images in a preferred embodiment of the present invention;
fig. 3 is a schematic structural diagram of a three-dimensional non-linear registration system for a large volume brain image data set according to a preferred embodiment of the present invention.
Detailed Description
The present invention is described in further detail below with reference to the attached drawing figures. Fig. 1 shows a flowchart of a three-dimensional non-linear registration method for a mass brain image data set according to a preferred embodiment of the present invention.
A three-dimensional non-linear registration method for a massive brain image data set comprises the following steps:
s0., a down-sampling step, which is to down-sample the preset high-resolution reference image data and the high-resolution image data to be registered to obtain the low-resolution reference image data and the low-resolution image data to be registered.
Wherein the high resolution reference image data D1, the high resolution image data to be registered D2. And performing down-sampling on the two high-resolution image data to obtain low-resolution reference image data D3 and low-resolution image data D4 to be registered. The down-sampling process is guaranteed by two points: the three-dimensional resolutions of the low-resolution image data in all directions are consistent; the two low-resolution image data D3, D4 have the same overall resolution.
S1, a low-resolution image registration step, which comprises the following steps:
s11, a linear registration step, namely performing linear registration on the low-resolution reference image data and the low-resolution image data to be registered to obtain a transformation matrix of the low-resolution linear registration, and calculating an inverse matrix of the low-resolution linear registration and a result after the low-resolution linear registration;
linear registration is carried out on the low-resolution image data D3 and D4 by using linear registration based on affine transformation to obtain a transformation matrix M1 of the low-resolution linear registration, and an inverse matrix M2 of the inverse matrix of the low-resolution linear registration and a result D5 after the low-resolution linear registration are calculated.
The linear registration in step S11 is all linear registration methods capable of describing the transformation mapping as a transformation matrix, such as: translation, rotation, scaling, etc.
S12, a nonlinear registration step, namely performing nonlinear registration on the result after the low-resolution linear registration and the low-resolution reference image data to obtain a displacement field of the low-resolution nonlinear registration and a result after the low-resolution nonlinear registration;
and carrying out displacement field-based nonlinear registration on the low-resolution linear registration result D5 and the low-resolution reference image data D3 to obtain a displacement field T1 of the low-resolution nonlinear registration and a low-resolution nonlinear registration result.
The non-linear registration in step S12 is all non-linear registration methods that can describe the transformation mapping as a displacement field of pixel offset vectors, such as: non-linear registration of elastic and fluid model transformations
S2, a high resolution image fast transformation step, which comprises the following steps:
s21, calculating a transformation matrix of high-resolution registration, namely calculating the transformation matrix of high-resolution registration and an inverse matrix of inverse matrix high-resolution registration based on the transformation matrix of low-resolution linear registration;
wherein, the transformation matrix M3 of high resolution registration and the inverse matrix M4 of inverse matrix high resolution registration are calculated based on the transformation matrix M1 of low resolution linear registration, and the formula is as follows:
M3=scaleM×M1×scaleM′
M4=M3-1
wherein, scaleM and scaleM' are respectively a reduction matrix and a magnification matrix, and the reduction and magnification times are consistent with the high resolution and the low resolution times.
S22, calculating a space range after high-resolution registration, namely calculating the space range after high-resolution registration by using the inverse matrix of high-resolution registration obtained in the step S21;
wherein, 8 vertexes of the high-resolution image data D2 to be registered are transformed by using the high-resolution inverse matrix M4, and the maximum and minimum values in the X, Y, Z direction are taken as the high-resolution space range after registration of the high-resolution image data D6.
S23, a blocking step, namely blocking the high-resolution space range obtained in the step S22 after registration;
wherein, the block is divided based on the computing resource, and the transformation of the pixel points in the block is calculated in parallel. And partitioning the space range after the high-resolution registration in the Z direction according to the computing environment. If the computing environment can support n threads in parallel at maximum, blocking is done with length n, which enables parallel computation of the same piece of data. If the memory data to be loaded into each block exceeds the memory, the block length is reduced by half until the memory requirement is met.
And S24, a high-resolution nonlinear transformation step, namely performing parallel computation on each block obtained in the step S23, computing the corresponding position of each pixel point in the original high-resolution image data to be registered, and finally obtaining a high-resolution image nonlinear transformation result.
The high resolution nonlinear transformation described in the step S24 includes the following sub-steps:
s241, estimating the range of the original high-resolution image data to be registered before block corresponding transformation;
and S242, traversing each pixel point in the block, calculating an offset vector corresponding to the pixel point through interpolation based on a displacement field of low-resolution nonlinear registration, calculating a corresponding position of the pixel point in original high-resolution image data to be registered based on the offset vector and an inverse matrix of high-resolution registration, and taking a gray value of the pixel point as a gray value of the pixel point.
Wherein, for each block, the Z-direction range of the original high-resolution image data D2 to be registered is calculated first. Calculating the block of the low-resolution image data D4 to be registered corresponding to the block, adding the offset vectors of the displacement field T1 corresponding to the low-resolution non-linear registration in three directions to the position of each pixel in the block, and then multiplying the offset vectors by the inverse matrix M2 of the low-resolution linear registration to obtain the position of the original low-resolution image data D4 to be registered corresponding to the block. After the positions of all the pixel points of the blocks corresponding to the original low-resolution image data D4 to be registered are solved, the maximum and minimum values in three directions are calculated to obtain the Z-direction range of the original low-resolution image data D4 to be registered, and finally the Z-direction range corresponding to the original high-resolution image data D2 to be registered is calculated.
As shown in fig. 2. For a pixel point p0(x0,y0,z0) Calculating its coordinate pixel p in the low-resolution image data D4 to be registered1(x1,y1,z1) Then its coordinate corresponding to the displacement field T1 of the low resolution non-linear registration is (x)1,y1,z1) Due to x1,y1,z1Not necessarily an integer, so that it is necessary to perform trilinear interpolation on the corresponding position of the displacement field T1 of the low-resolution nonlinear registration to calculate the pixel point p1Corresponding offset vector d1. According to calculated p1Dot offset vector d1Calculate p0Corresponding offset vector d0. To pixel point p0Addition of coordinate values to offset vectors d0The obtained coordinates p2Then multiplying by an inverse matrix M4 of high-resolution registration to finally obtain a coordinate p3Is p0Coordinates of inverse transformation, i.e. pixel point p0Inversely transforming the corresponding point in the original high-resolution image data D2 to be registered, and taking the gray value of the point as p0The gray value of (a).
And calculating the gray value corresponding to each pixel in the block, and writing each segment of gray value into a magnetic disk in a two-dimensional image form after the calculation is finished. And circularly calculating each block until the whole high-resolution image data to be registered is calculated.
From the above, the present invention utilizes the low-resolution image data registration information to transform the massive high-resolution three-dimensional image data set, thereby realizing the nonlinear registration of the massive high-resolution three-dimensional image data set. The method is suitable for all linear registration based on the transformation matrix and all nonlinear registration based on the displacement field, and can rapidly carry out nonlinear registration on the TB-level data set.
Referring to fig. 3, a schematic structural diagram of a three-dimensional non-linear registration system for a mass brain image data set according to an embodiment of the present invention is shown.
A system for three-dimensional non-linear registration of a set of massive brain image data, comprising the following modules:
the down-sampling module is used for down-sampling preset high-resolution reference image data and high-resolution image data to be registered to obtain low-resolution reference image data and low-resolution image data to be registered;
a low resolution image registration module, comprising:
the linear registration unit is used for carrying out linear registration on the low-resolution reference image data and the low-resolution image data to be registered to obtain a transformation matrix of the low-resolution linear registration, and calculating an inverse matrix of the low-resolution linear registration and a result after the low-resolution linear registration;
the nonlinear registration unit is used for carrying out nonlinear registration on the low-resolution linear registration result and the low-resolution reference image data to obtain a displacement field of the low-resolution nonlinear registration and a low-resolution nonlinear registration result;
the high resolution image fast transformation module comprises:
the high-resolution registration transformation matrix calculation unit is used for calculating a high-resolution registration transformation matrix and an inverse matrix of the inverse matrix high-resolution registration transformation matrix based on the low-resolution linear registration transformation matrix;
the high-resolution post-registration spatial range calculating unit calculates a high-resolution post-registration spatial range by using the inverse matrix of the high-resolution registration obtained by the high-resolution registration transformation matrix calculating unit;
the blocking unit is used for blocking the high-resolution post-registration spatial range obtained by the high-resolution post-registration spatial range calculating unit;
and the high-resolution nonlinear transformation unit is used for performing parallel calculation on each block obtained by the block dividing unit, calculating the corresponding position of each pixel point in the original high-resolution image data to be registered, and finally obtaining a high-resolution image nonlinear transformation result.
The linear registration in the linear registration unit describes the transformation mapping as a transformation matrix. The non-linear registration, described in the non-linear registration unit, describes the transformation mapping as a displacement field of pixel offset vectors.
In the blocking unit, the blocking is performed based on computing resources, the transformation of pixel points in the blocking is calculated in parallel, the computing resources can support n threads to be parallel at most, and the blocking is performed according to the length n.
The high resolution nonlinear transformation in the high resolution nonlinear transformation unit comprises the following units: the estimation unit estimates the range of the original high-resolution image data to be registered before the block corresponding transformation; and the transformation unit is used for traversing each pixel point in the blocks, calculating an offset vector corresponding to the pixel point through interpolation based on the displacement field of low-resolution nonlinear registration, calculating the corresponding position of the pixel point in the original high-resolution image data to be registered based on the offset vector and the inverse matrix of high-resolution registration, and taking the gray value of the pixel point as the gray value of the pixel point.
From the above, the present invention utilizes the low-resolution image data registration information to transform the massive high-resolution three-dimensional image data set, thereby realizing the nonlinear registration of the massive high-resolution three-dimensional image data set. The method is suitable for all linear registration based on the transformation matrix and all nonlinear registration based on the displacement field, and can rapidly carry out nonlinear registration on the TB-level data set.
Finally, it should be noted that the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, which should be covered by the claims of the present invention.
Claims (8)
1. A three-dimensional non-linear registration method for a massive brain image dataset is characterized by comprising the following steps:
s0, a down-sampling step, namely, down-sampling preset high-resolution reference image data and high-resolution image data to be registered to obtain low-resolution reference image data and low-resolution image data to be registered;
s1, a low resolution image registration step, which includes:
s11, a linear registration step, namely performing linear registration on the low-resolution reference image data and the low-resolution image data to be registered to obtain a transformation matrix of the low-resolution linear registration, and calculating an inverse matrix of the low-resolution linear registration and a result after the low-resolution linear registration;
s12, a nonlinear registration step, namely performing nonlinear registration on the result after the low-resolution linear registration and the low-resolution reference image data to obtain a displacement field of the low-resolution nonlinear registration and a result after the low-resolution nonlinear registration;
s2, high resolution image fast transformation step, which includes:
s21, calculating a transformation matrix of high-resolution registration, namely calculating the transformation matrix of high-resolution registration and an inverse matrix of inverse matrix high-resolution registration based on the transformation matrix of low-resolution linear registration;
s22, calculating a space range after high-resolution registration, namely calculating the space range after high-resolution registration by using the inverse matrix of high-resolution registration obtained in the step S21;
s23, a blocking step, namely blocking the high-resolution space range obtained in the step S22 after registration;
s24, a high-resolution nonlinear transformation step, namely performing parallel computation on each block obtained in the step S23, and computing the corresponding position of each pixel point in the original high-resolution image data to be registered to finally obtain a high-resolution image nonlinear transformation result;
the high resolution nonlinear transformation described in the step S24 includes the following sub-steps:
s241, estimating the range of the original high-resolution image data to be registered before block corresponding transformation;
and S242, traversing each pixel point in the block, calculating an offset vector corresponding to the pixel point through interpolation based on a displacement field of low-resolution nonlinear registration, calculating a corresponding position of the pixel point in original high-resolution image data to be registered based on the offset vector and an inverse matrix of high-resolution registration, and taking a gray value of the pixel point as a gray value of the pixel point.
2. The method for three-dimensional non-linear registration of a large volume of brain image data set according to claim 1, wherein said linear registration in step S11 describes the transformation mapping as a transformation matrix.
3. The method for three-dimensional non-linear registration of a large volume of brain image data set according to claim 1, wherein said non-linear registration in step S12 describes the transformation mapping as a displacement field of pixel shift vectors.
4. The method for three-dimensional nonlinear registration of a large amount of brain image data sets according to claim 1, wherein the blocking in step S23 is performed based on computing resources, and transformation of pixel points in the blocking is performed in parallel, and the computing resources can support n threads at maximum to perform parallel processing, and the blocking is performed with a length of n.
5. A system for three-dimensional non-linear registration of a set of mass brain image data, comprising the following modules:
the down-sampling module is used for down-sampling preset high-resolution reference image data and high-resolution image data to be registered to obtain low-resolution reference image data and low-resolution image data to be registered;
a low resolution image registration module, comprising:
the linear registration unit is used for carrying out linear registration on the low-resolution reference image data and the low-resolution image data to be registered to obtain a transformation matrix of the low-resolution linear registration, and calculating an inverse matrix of the low-resolution linear registration and a result after the low-resolution linear registration;
the nonlinear registration unit is used for carrying out nonlinear registration on the low-resolution linear registration result and the low-resolution reference image data to obtain a displacement field of the low-resolution nonlinear registration and a low-resolution nonlinear registration result;
the high resolution image fast transformation module comprises:
the high-resolution registration transformation matrix calculation unit is used for calculating a high-resolution registration transformation matrix and an inverse matrix of the inverse matrix high-resolution registration transformation matrix based on the low-resolution linear registration transformation matrix;
the high-resolution post-registration spatial range calculating unit calculates a high-resolution post-registration spatial range by using the inverse matrix of the high-resolution registration obtained by the high-resolution registration transformation matrix calculating unit;
the blocking unit is used for blocking the high-resolution post-registration spatial range obtained by the high-resolution post-registration spatial range calculating unit;
the high-resolution nonlinear transformation unit is used for performing parallel calculation on each block obtained by the block dividing unit, calculating the corresponding position of each pixel point in the original high-resolution image data to be registered, and finally obtaining a high-resolution image nonlinear transformation result;
the high resolution nonlinear transformation in the high resolution nonlinear transformation unit comprises the following units:
the estimation unit estimates the range of the original high-resolution image data to be registered before the block corresponding transformation;
and the transformation unit is used for traversing each pixel point in the blocks, calculating an offset vector corresponding to the pixel point through interpolation based on the displacement field of low-resolution nonlinear registration, calculating the corresponding position of the pixel point in the original high-resolution image data to be registered based on the offset vector and the inverse matrix of high-resolution registration, and taking the gray value of the pixel point as the gray value of the pixel point.
6. The system for three-dimensional non-linear registration of a large volume of brain image data set according to claim 5, wherein said linear registration in said linear registration unit describes a transformation mapping as a transformation matrix.
7. The system for three-dimensional non-linear registration of a large volume of brain image data set according to claim 5, wherein said non-linear registration in said non-linear registration unit describes a transformation mapping as a displacement field of pixel offset vectors.
8. The system according to claim 5, wherein said blocks in said blocking unit are blocked based on computing resources, and transformation of pixel points in said blocks is computed in parallel, said computing resources supporting at most n threads in parallel, and blocking with a length of n.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810307540.2A CN108537723B (en) | 2018-04-08 | 2018-04-08 | Three-dimensional nonlinear registration method and system for massive brain image data sets |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810307540.2A CN108537723B (en) | 2018-04-08 | 2018-04-08 | Three-dimensional nonlinear registration method and system for massive brain image data sets |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108537723A CN108537723A (en) | 2018-09-14 |
CN108537723B true CN108537723B (en) | 2021-09-28 |
Family
ID=63483358
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810307540.2A Active CN108537723B (en) | 2018-04-08 | 2018-04-08 | Three-dimensional nonlinear registration method and system for massive brain image data sets |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108537723B (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11238591B1 (en) | 2020-07-15 | 2022-02-01 | Taipei Medical University (Tmu) | Medical image processing system and method thereof |
TWI737404B (en) * | 2020-07-15 | 2021-08-21 | 臺北醫學大學 | Medical image processing system and method thereof |
CN112132877A (en) * | 2020-09-30 | 2020-12-25 | 东南大学 | Image registration improvement algorithm based on SIFT (scale invariant feature transform) features |
CN113516693B (en) * | 2021-05-21 | 2023-01-03 | 郑健青 | Rapid and universal image registration method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1883386A (en) * | 2006-07-06 | 2006-12-27 | 上海交通大学 | Elastic registration method of stereo MRI brain image based on machine learning |
CN101238993A (en) * | 2008-02-01 | 2008-08-13 | 哈尔滨工业大学 | Medical ultrasound image registration method based on integer lifting wavelet multiresolution analysis |
CN104933716A (en) * | 2015-06-16 | 2015-09-23 | 山东大学(威海) | Non-rigid registration method applied to medical image |
CN105931184A (en) * | 2016-04-18 | 2016-09-07 | 西安电子科技大学 | SAR image super-resolution method based on combined optimization |
CN106803267A (en) * | 2017-01-10 | 2017-06-06 | 西安电子科技大学 | Indoor scene three-dimensional rebuilding method based on Kinect |
US9858640B1 (en) * | 2015-07-15 | 2018-01-02 | Hrl Laboratories, Llc | Device and method for merging 3D point clouds from sparsely distributed viewpoints |
-
2018
- 2018-04-08 CN CN201810307540.2A patent/CN108537723B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1883386A (en) * | 2006-07-06 | 2006-12-27 | 上海交通大学 | Elastic registration method of stereo MRI brain image based on machine learning |
CN101238993A (en) * | 2008-02-01 | 2008-08-13 | 哈尔滨工业大学 | Medical ultrasound image registration method based on integer lifting wavelet multiresolution analysis |
CN104933716A (en) * | 2015-06-16 | 2015-09-23 | 山东大学(威海) | Non-rigid registration method applied to medical image |
US9858640B1 (en) * | 2015-07-15 | 2018-01-02 | Hrl Laboratories, Llc | Device and method for merging 3D point clouds from sparsely distributed viewpoints |
CN105931184A (en) * | 2016-04-18 | 2016-09-07 | 西安电子科技大学 | SAR image super-resolution method based on combined optimization |
CN106803267A (en) * | 2017-01-10 | 2017-06-06 | 西安电子科技大学 | Indoor scene three-dimensional rebuilding method based on Kinect |
Non-Patent Citations (4)
Title |
---|
3D Affine Registration of Large Image Stack for High-Resolution Brain Imaging Techniques;Chaozhen Tan 等;《14th International Conference on Photonics and Imaging in Biology and Medicine》;20170928;第1-3页 * |
A New Coarse-to-Fine Framework for 3D Brain MR Image Registration;Terrence Chen 等;《International Workshop on Computer Vision for Biomedical Image Applications》;20051231;第114-124页 * |
Evaluation of 14 nonlinear deformation algorithms applied to human brain MRI registration;Arno Klein 等;《NeuroImage》;20090113;第46卷;第786-802页 * |
基于B样条自由形变三维医学图像非刚性配准研究;冯兆美 等;《影像科学与光化学》;20140315;第32卷(第2期);第200-208页 * |
Also Published As
Publication number | Publication date |
---|---|
CN108537723A (en) | 2018-09-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108537723B (en) | Three-dimensional nonlinear registration method and system for massive brain image data sets | |
CN107330439B (en) | Method for determining posture of object in image, client and server | |
US11593654B2 (en) | System for performing convolutional image transformation estimation | |
CN110637461B (en) | Compact optical flow handling in computer vision systems | |
CN110992243B (en) | Intervertebral disc cross-section image construction method, device, computer equipment and storage medium | |
CN111914756A (en) | Video data processing method and device | |
Hauenstein et al. | Adaptive strategies for solving parameterized systems using homotopy continuation | |
CN113393468A (en) | Image processing method, model training device and electronic equipment | |
US9129398B2 (en) | Edgel sampling for edge-based tracking | |
CN112949576B (en) | Attitude estimation method, apparatus, device and storage medium | |
CN110651475B (en) | Hierarchical data organization for compact optical streaming | |
CN117541632A (en) | Multi-mode image registration method based on feature enhancement and multi-scale correlation | |
CN117671031A (en) | Binocular camera calibration method, device, equipment and storage medium | |
CN111161330B (en) | Non-rigid image registration method, device, system, electronic equipment and storage medium | |
CN115937002B (en) | Method, apparatus, electronic device and storage medium for estimating video rotation | |
CN106651950B (en) | Single-camera pose estimation method based on quadratic curve perspective projection invariance | |
CN103873875A (en) | Layering sub pixel motion estimation method for image super resolution | |
CN113591542B (en) | Visual servo control method, device and equipment for robot | |
CN117057982B (en) | Integrated circuit for performing distortion transformation on image and electronic equipment | |
CN114750147B (en) | Space pose determining method and device of robot and robot | |
JP7418517B2 (en) | Text recognition methods, devices, electronic devices, storage media and computer programs | |
CN115131401B (en) | Remote sensing video target tracking method based on multi-scale multi-direction kernel correlation filtering | |
Huang et al. | Non-uniform Image Down-sampling for Efficient Dense Perception | |
CN113838138B (en) | System calibration method, system, device and medium for optimizing feature extraction | |
Jammal et al. | Multi-resolution for disparity estimation with convolutional neural networks |
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 |