Diffeomorphism-based cross-modality brain region image registration method

Abstract

This invention relates to the field of biomedical image processing technology, and particularly to a cross-modal brain region image registration method based on differential homeomorphism. The method includes: registering the original T1-w image with an NMT-averaged standard template; registering the acquired cell architecture imaging and fluorescence imaging with the registered T1-w image; downsampling the fluorescence imaging and cell architecture imaging respectively; performing intensity correction; smoothing the cell architecture imaging / fluorescence imaging; re-downsampling the smoothed cell architecture imaging / fluorescence imaging; registering the T1-w image to the intensity-corrected cell architecture imaging / fluorescence imaging through affine transformation to obtain a deformation field; registering the T1-w D99 brain atlas to the cell architecture imaging / fluorescence imaging using the deformation field; and sampling the D99 brain atlas from the cell architecture imaging / fluorescence imaging to complete the registration. The advantages are: stronger robustness, higher accuracy, and no reliance on subsequent manual correction.

Description

Technical Field

[0001] This invention relates to the field of biomedical image processing technology, and in particular to a cross-modal brain region image registration method based on differential homeomorphism. Background Technology

[0002] Cellular architecture imaging and fluorescence imaging are optical staining imaging techniques used to study the structure of biological tissues and cells. In recent years, these two imaging techniques have been frequently used to study brain connectivity and track nerve fibers. Therefore, this places higher demands on multimodal registration and brain region segmentation. The combination of fluorescence staining methods with fluorescence microscopy has enabled fluorescence imaging to be widely used in the detection of bioactive substances and cell imaging, with cell imaging being an important research tool in the field of life sciences. Compared with other techniques, fluorescence staining has advantages such as high sensitivity, strong selectivity, and simple operation, and is currently a highly sensitive visualization and analysis technique widely used in cell imaging.

[0003] Currently, the most commonly used multimodal registration methods include mutual information registration, point cloud registration, and morphological feature-based registration. These methods have achieved good results on mouse brains. However, monkey brains have more complex brain structures and greater individual differences, and the deformation caused by brain slicing is also greater. These factors lead to many cortical areas and some internal brain structures being unable to be accurately registered when using the above registration methods, further resulting in inaccurate brain region segmentation and affecting the accuracy of subsequent analysis results. There are also multimodal registration methods using deep learning, which mainly generate images with similar intensity distributions to the registered images through style transfer, thereby improving the accuracy of registration. However, these deep learning models are based on 2D images, and 3D images need to be sliced ​​into 2D images before training. This method results in the loss of the upper and lower slice information of the 3D image, leading to inaccurate 2D image structure. Furthermore, the 3D image reconstructed from the 2D image generated by the model produces many artifacts and irrelevant background signals. Some methods also correct registration errors through manual correction. However, for high-resolution images, manual correction may be time-consuming and requires operators to have a certain level of anatomical knowledge.

[0004] In summary, to address the shortcomings of current multimodal registration methods for monkey brains, an automated and precise method is needed. This precise multimodal registration would provide high-quality and accurate brain maps for mesoscopic biological analysis, brain connectivity, and neural fiber tracing. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a cross-modal brain region image registration method based on differential homeomorphism.

[0006] The purpose of this invention is to provide a cross-modal brain region image registration method based on differential homeomorphism, comprising the following steps:

[0007] S1. Register the original T1-w image with the average standard template of the neuromorphic tracing atlas of the macaque brain to obtain the registered T1-w image;

[0008] S2. Register the acquired cell architecture imaging and fluorescence imaging with the registered T1-w images; specifically including the following sub-steps:

[0009] S201. Downsample fluorescence imaging and cell architecture imaging respectively;

[0010] S202. Intensity correction is applied to cell architecture imaging / fluorescence imaging.

[0011] S203. Smooth the cell architecture imaging / fluorescence imaging;

[0012] S204. Resample the smoothed cell architecture imaging / fluorescence imaging;

[0013] S205. Register the T1-w image onto intensity-corrected cell architecture imaging / fluorescence imaging using affine transformation to obtain the deformation field;

[0014] S206. Using a deformation field, the D99 brain map of T1-w is registered onto cell architecture imaging / fluorescence imaging.

[0015] S207. Sample from the D99 brain atlas on cell construction imaging / fluorescence imaging to complete registration.

[0016] Preferably, the downsampling in step S201 involves downsampling the fluorescence imaging and cell architecture imaging to 0.075*0.075*0.075mm. 3 In step S204, the re-sampling involves downsampling the smoothed cell architecture imaging / fluorescence imaging to 0.25*0.25*0.25mm. 3 In step S207, the resolution after upsampling is 0.075*0.075*0.075mm. 3 .

[0017] Preferably, the affine transformation registration in step S205 is a differential homeomorphic nonlinear registration.

[0018] Preferably, the CLAHE algorithm is used for intensity correction in step S202; and Rice distribution is used for smoothing in step S203.

[0019] Preferably, step S1 specifically includes the following sub-steps:

[0020] S101. Perform an affine transformation on the T1-w image to adjust the orientation of the brain on the T1-w image;

[0021] S102. Perform affine registration between the T1-w image with the adjusted orientation and the NMT average standard template of the neuromorphic tracing standard atlas of the macaque brain, and align them with the AC-PC axis.

[0022] S103. Perform nonlinear registration of the aligned T1-w image with the NMT average standard template using differential homeomorphism, and obtain the registered deformation field.

[0023] S104. Perform bias field correction on T1-w after skull removal;

[0024] S105. After correction, the T1-w images of the brain are again nonlinearly registered with the NMT average standard template using differential homeomorphism, and the deformation field is obtained.

[0025] S106. By inversely transforming the deformation field, the brain map D99 in the NMT space is registered onto the T1-w image to obtain the registered T1-w image.

[0026] Preferably, the nonlinear registration of the differential homeomorphism in steps S103 and S105 uses mutual information and mean square error as evaluation indicators.

[0027] The preferred expression for the mean squared error is as follows:

[0028]

[0029] Where M is the total number of pixels in image I, N is the total number of pixels in image K, i represents the row of the image, j represents the column of the image, and the values ​​of i and j are both integers.

[0030] Preferably, the nonlinear registration of the differential homeomorphism in step S103 includes the following steps:

[0031] S1031. By inversely transforming the deformation field, the brain mask on the NMT average standard template space is registered onto the T1-w image in the individual space.

[0032] S1032. Use a mask to remove the skull and obtain a T1-w image with only brain tissue removed from the skull.

[0033] Preferably, the bias field correction method in step S104 uses the N4 bias field correction of ANTs.

[0034] Preferably, in step S2, the fMOST imaging system is used to acquire cell architecture imaging and fluorescence imaging, wherein the cell architecture imaging and fluorescence imaging are two channels acquired by the fMOST imaging system.

[0035] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0036] (1) This invention is a fully automatic atlas registration method with high accuracy and does not rely on manual correction in the later stage, thus solving the shortcomings of relying on human anatomical experience;

[0037] (2) This invention is based on the differential homeomorphism mapping, which is a smooth and reversible mapping that can ensure the structural similarity between processed images and prevent distortion. Compared with the traditional method that uses small deformation model and spatial transformation composite algorithm, the differential homeomorphism method has stronger robustness and more accurate results.

[0038] (3) Generally, nonlinear registration uses a single evaluation index, which can achieve good results when processing single-modality or high-quality images. However, for multimodal, low-contrast, or large-scale image structure differences, the results do not meet expectations. This method uses multiple evaluation indexes and assigns different weights based on the characteristics of different evaluation indexes, thereby improving the accuracy of the results. Attached Figure Description

[0039] Figure 1 This is a flowchart of the registration process between an individual T1-w image and a standard template NMT, provided according to an embodiment of the present invention.

[0040] Figure 2 This is a schematic diagram of the brain orientation before and after adjustment on T1-w images provided in an embodiment of the present invention.

[0041] Figure 3 The images are T1w images (A) and NMT standard templates (B) after differential homeomorphic registration according to embodiments of the present invention.

[0042] Figure 4 The deformation field results provided according to the embodiments of the present invention are: (A) the deformation field generated by T1-w registration to NMT; (B) the deformation field after inverse transformation.

[0043] Figure 5 These are brain mask images provided according to embodiments of the present invention; (A) a brain mask of T1-w; (B) a brain mask of the standard template NMT.

[0044] Figure 6 The image shown is a T1-w image of a macaque brain provided according to an embodiment of the present invention.

[0045] Figure 7 The results are based on the brain atlas D99 provided in the embodiments of the present invention in the T1-w(A) and NMT average standard template (B) spaces.

[0046] Figure 8These are downsampling results of fluorescence imaging (A) and cell architecture imaging (B) provided according to embodiments of the present invention.

[0047] Figure 9 This is an image of the D99 brain atlas in PI space provided according to an embodiment of the present invention.

[0048] Figure 10 This is a flowchart of a cross-modal brain region image registration method based on differential homeomorphism provided in an embodiment of the present invention. Detailed Implementation

[0049] In the following description, embodiments of the invention will be described with reference to the accompanying drawings. In the description below, the same modules are denoted by the same reference numerals. Where the same reference numerals are used, their names and functions are also the same. Therefore, their detailed description will not be repeated.

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0051] This invention provides a cross-modal brain region image registration method based on differential homeomorphism, comprising the following steps:

[0052] S1. Registration of the original T1-w image with the NMT average standard template of the neuromorphic tracing atlas of the macaque brain, specifically including the following sub-steps:

[0053] S101. Perform an affine transformation on the T1-w image to adjust the orientation of the brain in the T1-w image. Since MRI equipment is designed specifically for humans, and the body and head shape of monkeys differ significantly from those of humans, the orientation of the head is inconsistent with the standard space NMT when photographing the monkey brain. Through coordinate transformation, the orientation can be adjusted more effectively for subsequent processing.

[0054] S102. Perform affine registration between the T1-w image with the adjusted orientation and the NMT average standard template of the neuromorphic tracing standard atlas of the macaque brain, aligning them with the AC-PC axes.

[0055] S103, such as Figure 3 As shown, mutual information (MI) and mean square error (MSE) are used as evaluation metrics. The aligned T1-w image is nonlinearly registered with the NMT average standard template using differential homeomorphism, and the deformation field after registration is obtained. Figure 4 );

[0056] The mean square error expression is as follows:

[0057]

[0058] Where M is the total number of pixels in image I, N is the total number of pixels in image K, i represents the row of the image, j represents the column of the image, and the values ​​of i and j are both integers.

[0059] The smaller the MSE value, the more similar the images are.

[0060] Specifically, the nonlinear registration of differential homeomorphisms includes the following steps:

[0061] S1031. By inversely transforming the deformation field, the brain mask in the NMT average standard template space is registered onto the T1-w image in the individual space. Figure 5 );

[0062] S1032. Use a mask to remove the skull and obtain a T1-w image containing only brain tissue. Figure 6 ).

[0063] Principle: Nonlinear registration typically uses a single evaluation metric, which yields good results when processing single-modality or high-quality images. However, for multimodal images, images with low contrast, or images with significant structural differences, the results are unsatisfactory. Using multiple evaluation metrics and assigning different weights based on their characteristics can improve the results. Mutual information (MI), derived from information theory, is a classic similarity measure used for image registration, generally applied to multimodal image registration. Although the T1-w image and the standard template NMT belong to the same modality, they have significant intensity differences. Higher weighted mutual information can achieve better registration results. MSE is one of the metrics for measuring image quality, calculated by summing and averaging the squares of the differences between the true and predicted values. Traditional methods utilize small deformation models, and spatial transformation composite algorithms use vector addition for approximation. However, small deformations can cause local folding, destroying the original topological image. This method uses a differential homeomorphic mapping, which is smooth and invertible and can guarantee the structural similarity between processed images, preventing distortion.

[0064] S104. Using the N4 bias field correction method of ANTs, bias field correction is performed on T1-w after skull removal.

[0065] S105. Using mutual information and mean square error as evaluation indicators, the T1-w images of the brain are again nonlinearly registered with the NMT average standard template by differential homeomorphism, and the deformation field is obtained.

[0066] After removing the skull, performing nonlinear registration using differential homeomorphism again can remove redundant skull information, improve the registration accuracy of the cerebral sulci and gray-white matter junctions, and make the subsequent D99 brain atlas registration more accurate.

[0067] S106. By inversely transforming the deformation field, the brain map D99 in NMT space is registered onto the T1-w image. Figure 7 ).

[0068] S2. Register the acquired cell architecture imaging (PI), fluorescence imaging (GFP), and T1-w images, specifically including the following sub-steps:

[0069] S201. Downsample fluorescence imaging and cell architecture imaging to 0.075*0.075*0.075mm respectively. 3 ( Figure 8 );

[0070] S202. Intensity correction was performed on cell architecture imaging / fluorescence imaging using the CLAHE algorithm.

[0071] S203. Smooth cell architecture imaging / fluorescence imaging using the Rician distribution;

[0072] S204. Downsample the smoothed cell architecture imaging / fluorescence imaging to 0.25*0.25*0.25mm. 3 ;

[0073] S205, 0.5*0.5*0.5mm 3 The T1-w image was registered to 0.25*0.25*0.25mm using an affine transformation. 3 Deformation fields were obtained on intensity-corrected cell architecture imaging / fluorescence imaging; specifically, affine transformation registration was converted to differential homeomorphic nonlinear registration.

[0074] S206. Using a deformation field, the D99 brain map of T1-w was registered to a size of 0.25*0.25*0.25mm. 3 3. Cell architecture imaging / fluorescence imaging (on) Figure 9 ).

[0075] S207, will be 0.25*0.25*0.25mm 3 D99 brain atlases on cell architecture imaging / fluorescence imaging were upsampled to 0.075*0.075*0.075mm. 3 Resolution, registration complete.

[0076] Cellular construct imaging (PI) and fluorescence imaging (GFP) were acquired using the fMOST imaging system, which consists of two channels acquired by the fMOST imaging system. Cellular construct imaging provides clearer outlines of brain structures than fluorescence imaging, so registration methods can be performed on cellular construct imaging.

[0077] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0078] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A cross-modal brain region image registration method based on differential homeomorphism, characterized in that, Includes the following steps: S1. Register the original T1-w image with the average standard template of the neuromorphic tracing atlas of the macaque brain to obtain the registered T1-w image; specifically including the following sub-steps: S101. Perform an affine transformation on the T1-w image to adjust the orientation of the brain on the T1-w image; S102. Perform affine registration between the T1-w image with the adjusted orientation and the NMT average standard template of the neuromorphic tracing standard atlas of the macaque brain, and align them with the AC-PC axis. S103. By inversely transforming the deformation field, the brain mask on the NMT average standard template space is registered onto the T1-w image in the individual space. Using a mask to remove the skull, T1-w images were obtained showing only brain tissue with the skull removed. S104. Perform bias field correction on T1-w after skull removal; S105. After correction, the T1-w image of the brain is again nonlinearly registered with the NMT average standard template using differential homeomorphism to remove redundant skull information and obtain the deformation field. S106. By inversely transforming the deformation field, the brain map D99 in the NMT space is registered onto the T1-w image to obtain the registered T1-w image. S2. Register the cell architecture imaging and fluorescence imaging acquired by the fMOST imaging system with the registered T1-w images; the cell architecture imaging and fluorescence imaging are two channels acquired by the fMOST imaging system; specifically, this includes the following sub-steps: S201. Downsample fluorescence imaging and cell architecture imaging respectively; S202. Intensity correction is applied to cell architecture imaging / fluorescence imaging. S203. Smooth the cell architecture imaging / fluorescence imaging; S204. Resample the smoothed cell architecture imaging / fluorescence imaging; S205. The T1-w image is registered onto intensity-corrected cell architecture imaging / fluorescence imaging using affine transformation to obtain the deformation field; the affine transformation registration is differential homeomorphic nonlinear registration. S206. Using a deformation field, the D99 brain map of T1-w is registered onto cell architecture imaging / fluorescence imaging. S207. Sample from the D99 brain atlas on cell construction imaging / fluorescence imaging to complete registration.

2. The cross-modal brain region image registration method based on differential homeomorphism according to claim 1, characterized in that: The downsampling in step S201 involves downsampling the fluorescence imaging and cell construction imaging to 0.

075. 0.075 0.075 mm³; the re-sampling in step S204 involves downsampling the smoothed cell architecture imaging / fluorescence imaging to 0.25 mm³. 0.25 0.25mm³; the resolution after upsampling in step S207 is 0.

075. 0.075 0.075mm³.

3. The cross-modal brain region image registration method based on differential homeomorphism according to claim 2, characterized in that: In step S202, the CLAHE algorithm is used for intensity correction; in step S203, Rice distribution is used for smoothing.

4. The cross-modal brain region image registration method based on differential homeomorphism according to claim 1, characterized in that: The nonlinear registration of the differential homeomorphism in steps S103 and S105 uses mutual information and mean square error as evaluation indicators.

5. The cross-modal brain region image registration method based on differential homeomorphism according to claim 4, characterized in that: The mean square error expression is as follows: Where M is the total number of pixels in image I, N is the total number of pixels in image K, i represents the row of the image, j represents the column of the image, and the values ​​of i and j are both integers.

6. The cross-modal brain region image registration method based on differential homeomorphism according to claim 5, characterized in that: The bias field correction method in step S104 uses ANTs' N4 bias field correction.

Images