阅读说明：本技术 一种基于形状约束的连续脑片图像三维配准方法 (Continuous brain picture image three-dimensional registration method based on shape constraint ) 是由 丰钊 李安安 张邹涛 龚辉 骆清铭 于 2019-10-11 设计创作，主要内容包括：本发明提供了一种基于形状约束的连续脑片图像三维配准方法,涉及图像配准的技术领域。包括S1：将待配准的二维脑图像序列对齐,得到二维对齐序列,重建成初始三维体；S2：将标准脑图谱线性配准到初始三维体上,得到参考脑图谱,将二维对齐序列的图片逐张配准在参考脑图谱内对应位置的切面上,得到二维校正序列,重建成校正三维体；S3：获得校正三维体内各像素点的反向形变参数,根据反向形变参数对校正三维体进行空间偏移调整,得到精修三维体。把参考脑图谱作为全局形状的约束条件,从而将真实脑组织的正确形状引入到校正三维体重建中,解决了现有技术中由于缺少空间位置关系和全局形状信息而导致的轴向偏移问题。(The invention provides a three-dimensional registration method of continuous brain slice images based on shape constraint, and relates to the technical field of image registration. Including S1: aligning the two-dimensional brain image sequence to be registered to obtain a two-dimensional alignment sequence, and reconstructing into an initial three-dimensional body; s2: linearly registering the standard brain atlas to the initial three-dimensional body to obtain a reference brain atlas, registering pictures of the two-dimensional alignment sequence on a section of a corresponding position in the reference brain atlas one by one to obtain a two-dimensional correction sequence, and reconstructing into a corrected three-dimensional body; s3: and obtaining reverse deformation parameters of all pixel points in the corrected three-dimensional body, and carrying out spatial offset adjustment on the corrected three-dimensional body according to the reverse deformation parameters to obtain the refined three-dimensional body. The reference brain atlas is used as a constraint condition of the global shape, so that the correct shape of the real brain tissue is introduced into the corrected three-dimensional body reconstruction, and the problem of axial offset caused by lack of spatial position relation and global shape information in the prior art is solved.)

1. A three-dimensional registration method of continuous brain slice images based on shape constraint is characterized by comprising the following steps:

s1, local shape correction: aligning the pictures of the whole two-dimensional brain image sequence to be registered, obtaining a two-dimensional alignment sequence after all the pictures are aligned, and then reconstructing the two-dimensional alignment sequence into an initial three-dimensional body in a three-dimensional mode;

s2, global shape correction: linearly registering the standard brain atlas to the initial three-dimensional body to obtain a reference brain atlas, linearly registering pictures of the two-dimensional alignment sequence on a section of a corresponding position in the reference brain atlas one by one to obtain a two-dimensional correction sequence, and then performing three-dimensional reconstruction on the two-dimensional correction sequence to obtain a corrected three-dimensional body;

s3, three-dimensional shape finishing: and obtaining reverse deformation parameters of all pixel points in the corrected three-dimensional body, and carrying out spatial offset adjustment on the corrected three-dimensional body according to the reverse deformation parameters to obtain the refined three-dimensional body.

2. The method for three-dimensional registration of continuous brain slice images based on shape constraint according to claim 1, wherein the step S1 is specifically as follows:

s101, selecting one image from a two-dimensional brain image sequence to be registered as an initial image;

s102, starting from the initial picture, aligning two pictures adjacent to the initial picture on the initial picture one by one;

s103, sequentially aligning other pictures in the two-dimensional brain image sequence to be registered to the previous picture until the whole two-dimensional brain image sequence to be registered is traversed to obtain a two-dimensional alignment sequence;

and S104, performing three-dimensional reconstruction based on the two-dimensional alignment sequence to obtain an initial three-dimensional body.

3. The three-dimensional registration method for continuous brain slice images based on shape constraint according to claim 1, wherein in S2, the standard brain atlas is linearly registered to the initial three-dimensional body to obtain the reference brain atlas and a reference three-dimensional body containing the reference brain atlas data;

and then, by referring to the reference three-dimensional body, finding out the sections corresponding to the picture positions in the two-dimensional alignment sequence one by one in the reference brain atlas.

4. The three-dimensional registration method for continuous brain slice images based on shape constraint according to claim 3, wherein the linear transformation relation for linearly registering the standard brain atlas into the initial three-dimensional body in the step S2 is as follows:

T＝arg min MI(V _aligned,T(V _reference))；

wherein T represents a three-dimensional linear transformation parameter, V _alignedRepresenting an initial three-dimensional volume, V _referenceRepresenting a standard brain atlas and MI representing mutual information between the two.

5. The method for three-dimensional registration of continuous brain slice images based on shape constraint according to claim 3, wherein the linear transformation relationship based on which the two-dimensional alignment sequence is linearly registered one by one onto the tangent plane of the corresponding position of the reference brain atlas in S2 is:

I _i ^alignedi picture representing a two-dimensional aligned sequence, I _i ^referenceA section corresponding to the position of the reference brain atlas and the ith picture, MI refers to the mutual information between the two, T _iRepresents that _i ^alignedRegistration to I _i ^referenceThe two-dimensional linear transformation parameters of (1).

6. The method for three-dimensional registration of continuous brain slice images based on shape constraint according to claim 1, wherein the specific manner for obtaining the inverse deformation parameter in S3 is as follows:

the corrected three-dimensional body is non-linearly registered to a reference brain atlas in a three-dimensional space, and reverse deformation parameters of all pixel points are obtained;

the specific way of adjusting the spatial offset in S3 is as follows: and carrying out reverse nonlinear registration on the reference brain atlas to the corrected three-dimensional body according to the reverse deformation parameters.

7. The method for three-dimensional registration of continuous brain slice images based on shape constraint according to claim 6, wherein the non-linear registration in S3 requires obtaining a reverse deformation field by correcting gray scale and shape features of a three-dimensional body at different positions in a three-dimensional space, and the reverse deformation parameters are used for describing the reverse deformation field.

8. The method for three-dimensional registration of continuous brain slice images based on shape constraint according to claim 7, wherein the non-linear registration in S3 is as follows:

traversing all pixel points on the reference brain atlas, searching the deformation offset of the space coordinate in the reverse deformation field according to the space coordinate of the pixel point, and moving the space coordinate to a new position according to the corresponding deformation offset.

Technical Field

The invention relates to the technical field of image registration, in particular to a three-dimensional registration method of continuous brain slice images based on shape constraint.

Background

The brain atlas is often called a Baidu map in the fields of brain science, medicine and the like, and accurate spatial positioning information is provided for the work in different directions such as the development of clinical operation, accurate drug administration, cognitive behavior research, brain function loop research and the like by matching interested neurobiological signals such as neuron projection, cerebrovascular branches and the like into a standard coordinate system.

Constructing a fine brain atlas relies on higher spatial resolution image data. With the progress of micro-optical imaging technology, the currently acquired three-dimensional image data set can reach micron resolution level in the transverse direction and the axial direction, so that a three-dimensional high-precision brain atlas represented by a Common Coordinate Frame (CCF) is constructed, and a more accurate spatial information reference is provided for mesoscopic level neuroscience research.

At the same time, however, most laboratories today are limited in technology and capital, and still can only acquire neuro-image datasets using traditional sample preparation and imaging techniques. Such datasets are made up of a series of two-dimensional images spaced widely apart, with horizontal resolutions on the order of microns or even sub-microns, while axial spatial resolutions typically on the order of hundreds of microns, with two to three orders of magnitude difference. The large difference in horizontal and axial spatial resolution makes such datasets unsuitable for being considered as typical three-dimensional images, but only as two-dimensional image sequences, for which registration techniques established for three-dimensional images are therefore not applicable. For this reason, techniques have been developed specifically to localize such two-dimensional image sequences to three-dimensional brain atlases.

Disclosure of Invention

The invention aims to provide a continuous brain slice image three-dimensional registration method based on shape constraint, which aims to solve the problem of axial offset generated in the process of registering a two-dimensional brain image sequence to a three-dimensional brain atlas in the prior art.

A three-dimensional registration method of continuous brain slice images based on shape constraint comprises the following steps:

s1, local shape correction: aligning the pictures of the whole two-dimensional brain image sequence to be registered, obtaining a two-dimensional alignment sequence after all the pictures are aligned, and then reconstructing the two-dimensional alignment sequence into an initial three-dimensional body in a three-dimensional mode;

s2, global shape correction: linearly registering the standard brain atlas to the initial three-dimensional body to obtain a reference brain atlas, linearly registering pictures of the two-dimensional alignment sequence on a section of a corresponding position in the reference brain atlas one by one to obtain a two-dimensional correction sequence, and then performing three-dimensional reconstruction on the two-dimensional correction sequence to obtain a corrected three-dimensional body;

s3, three-dimensional shape finishing: and obtaining reverse deformation parameters of all pixel points in the corrected three-dimensional body, and carrying out spatial offset adjustment on the corrected three-dimensional body according to the reverse deformation parameters to obtain the refined three-dimensional body.

According to the technical scheme, the two-dimensional brain image sequence is aligned and reconstructed into an initial three-dimensional body, then a reference brain map is obtained by using a standard brain map and is used as a constraint reference of the overall shape, under the constraint condition, the two-dimensional aligned sequence obtained after alignment is registered on a tangent plane of a corresponding position in the reference brain map one by one, so that the correct shape of the real brain tissue is introduced into the reconstruction step of the corrected three-dimensional body, the data shape of the initial three-dimensional body is further optimized, and the overall shape of the obtained corrected three-dimensional body does not deviate from the correct shape. Finally, local and non-uniform out-of-plane deformation is corrected through spatial offset adjustment, so that the problem of axial offset caused by lack of spatial position relation and global shape information in the prior art is solved, and the final result can well accord with the overall shape of the real brain tissue.

Further, the S1 specifically includes:

s101, selecting one image from a two-dimensional brain image sequence to be registered as an initial image;

s102, starting from the initial picture, aligning two pictures adjacent to the initial picture on the initial picture one by one;

s103, sequentially aligning other pictures in the two-dimensional brain image sequence to be registered to the previous picture until the whole two-dimensional brain image sequence to be registered is traversed to obtain a two-dimensional alignment sequence;

and S104, performing three-dimensional reconstruction based on the two-dimensional alignment sequence to obtain an initial three-dimensional body.

Further, in the step S2, the standard brain atlas is linearly registered to the initial three-dimensional body, so that a reference brain atlas is obtained and a reference three-dimensional body containing reference brain atlas data is also obtained;

and then, by referring to the reference three-dimensional body, finding out the sections corresponding to the picture positions in the two-dimensional alignment sequence one by one in the reference brain atlas.

Further, in the step S2, the linear transformation relation according to which the standard brain atlas is linearly registered into the initial three-dimensional body is:

T＝argminMI(V _aligned,T(V _reference))；

wherein T represents a three-dimensional linear transformation parameter, V _alignedRepresenting an initial three-dimensional volume, V _referenceRepresenting a standard brain atlas and MI representing mutual information between the two.

Further, in S2, the linear transformation relation according to which the two-dimensional alignment sequence is linearly registered on the tangent plane at the corresponding position of the reference brain atlas one by one is:

I _i ^alignedi picture representing a two-dimensional aligned sequence, I _i ^referenceA section corresponding to the position of the reference brain atlas and the ith picture, MI refers to the mutual information between the two, T _iRepresents that _i ^alignedRegistration to I _i ^referenceThe two-dimensional linear transformation parameters of (1).

Further, the specific manner of acquiring the reverse deformation parameter in S3 is as follows:

the corrected three-dimensional body is non-linearly registered to a reference brain atlas in a three-dimensional space, and reverse deformation parameters of all pixel points are obtained;

the specific way of adjusting the spatial offset in S3 is as follows: and carrying out reverse nonlinear registration on the reference brain atlas to the corrected three-dimensional body according to the reverse deformation parameters.

Further, the non-linear registration in S3 requires that a reverse deformation field is obtained by correcting the gray scale and shape features of the three-dimensional body at different positions in the three-dimensional space, and the reverse deformation parameters are used for describing the reverse deformation field.

Further, the specific manner of the non-linear registration in S3 is as follows:

traversing all pixel points on the reference brain atlas, searching the deformation offset of the space coordinate in the reverse deformation field according to the space coordinate of the pixel point, and moving the space coordinate to a new position according to the corresponding deformation offset.

Drawings

FIG. 1 is a block flow diagram of the present invention;

fig. 2 a is a schematic diagram of a single picture in a two-dimensional correction sequence subjected to global shape correction;

fig. 2 b is a schematic diagram of a single picture in the two-dimensional correction sequence after the boundary information is mapped.

Detailed Description

In order to make the technical problems, technical solutions and advantageous effects to be solved by the present invention more clearly apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The embodiment specifically discloses a three-dimensional registration method of continuous brain slice images based on shape constraint, and with reference to fig. 1, the method comprises the following steps:

s1, local shape correction:

s101, selecting a picture with less damage from a two-dimensional brain image sequence to be registered as an initial picture.

And S102, starting from the initial picture, aligning two pictures adjacent to the initial picture on the initial picture one by applying a linear registration algorithm.

And S103, sequentially aligning other pictures in the two-dimensional brain image sequence to the previous picture until the whole two-dimensional brain image sequence is traversed, and finally obtaining a two-dimensional alignment sequence so as to eliminate local dislocation and deformation between adjacent pictures.

And S104, performing three-dimensional reconstruction based on the two-dimensional alignment sequence to obtain an initial three-dimensional body. The initial three-dimensional body is reconstructed after being aligned, the local shape is corrected, and the three-dimensional information which can be referred to in the subsequent steps is carried.

S2, global shape correction:

s201, a linear registration algorithm is applied to register the standard brain atlas on the initial three-dimensional body to obtain a reference brain atlas corresponding to the standard brain atlas, and a reference three-dimensional body containing reference brain atlas data is obtained at the same time, the reference three-dimensional body has the complete shape of each brain area in a three-dimensional space, the purpose of global shape correction is achieved, and the reference brain atlas can be combined to serve as reference constraint during subsequent processing.

The linear registration algorithm in this step uses a general optimization calculation method, and is denoted as T, T ═ argminMI (V) according to an optimal linear transformation relation _aligned,T(V _reference) In which T represents a three-dimensional linear transformation parameter, V) _alignedRepresenting an initial three-dimensional volume, V _referenceRepresenting a standard brain atlas and MI representing mutual information between the two.

S202, by referring to the reference three-dimensional body, one-by-one tangent plane corresponding to the position of the picture data in the two-dimensional alignment sequence is found in the reference brain atlas of S201, the pictures in the two-dimensional alignment sequence are registered to the corresponding tangent plane in the reference brain atlas one by using a linear registration algorithm, the registration of any two pictures is not interfered with each other, and the registration process can be expressed by the following formula:

wherein, I _i ^alignedI picture representing a two-dimensional aligned sequence, I _i ^referenceA section corresponding to the position of the reference brain atlas and the ith picture, MI refers to the mutual information between the two, T _iRepresents that _i ^alignedRegistration to I _i ^referenceThe two-dimensional linear transformation parameters of (1).

And S203, performing three-dimensional reconstruction on the two-dimensional correction sequence obtained in the S202 to obtain a corrected three-dimensional body.

In the linear registration process of S202, the two-dimensional alignment sequence can be matched with the shape of the corresponding tangent plane of the reference brain atlas under the shape constraint of the reference three-dimensional volume, thereby realizing global shape correction of all pictures in the two-dimensional alignment sequence, solving the overall shape deviation of the two-dimensional alignment sequence, and finally obtaining a corrected three-dimensional volume subjected to global shape correction.

S3, three-dimensional shape finishing:

s301, registering the corrected three-dimensional body obtained in the step S203 on the reference brain atlas in the step S201 in a three-dimensional space by using a nonlinear registration algorithm, and obtaining a reverse deformation parameter. The non-linear registration algorithm used in this step may be a Large Deformation difference Metric Mapping algorithm in ANTs TOOL, and this algorithm needs to obtain a reverse Deformation field by correcting the gray scale and shape characteristics of the three-dimensional body at different positions in the three-dimensional space, and the reverse Deformation parameters are used to describe the reverse Deformation field.

S302, performing spatial offset adjustment on the corrected three-dimensional body: and reversely and nonlinearly registering the reference brain atlas to the correction three-dimensional body according to the reverse deformation parameters, so that the boundary information of the reference brain atlas is matched to the correction three-dimensional body. The non-linear registration process is specifically as follows: traversing all pixel points on the reference brain atlas, searching deformation offset of the space coordinate in the reverse deformation field according to the space coordinate of the pixel point and the deformation parameter of the corresponding reverse deformation field when the pixel point corresponds to the reverse deformation field, and moving the space coordinate to a new position according to the corresponding deformation offset, so that the reference brain atlas is reversely and linearly registered to the corrected three-dimensional body, and the refined three-dimensional body is obtained. As shown in fig. 2, the three-dimensional boundary information of the brain region contained in the reference brain atlas can be mapped onto the two-dimensional correction sequence corresponding to the correction three-dimensional body, so as to be conveniently used for other analysis calculations into the analysis calculation of the neural image signal of interest.

The method comprises the steps of registering a standard brain atlas into an initial three-dimensional body to obtain a reference three-dimensional body and a reference brain atlas, using the reference three-dimensional body and the reference brain atlas as reference constraint conditions in the global shape correction and three-dimensional shape refinement processes, introducing the correct form of real brain tissue into the corrected three-dimensional body through sheet-by-sheet registration, and finally obtaining the refined three-dimensional body with accurate three-dimensional form information through space offset adjustment, so that the problem of axial offset is effectively avoided.

The above description is only a few preferred embodiments of the present invention, and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.