System and method for processing multiple rounds of three-dimensional biological tissue images and application

Abstract

The invention belongs to the field of biomedical image processing, provides a method and a system for processing multiple rounds of three-dimensional biological tissue images, comprises multiple rounds of three-dimensional image registration and three-dimensional cell segmentation, and further provides application of the method and the system in processing the multiple rounds of three-dimensional biological tissue images. According to the invention, through multi-step registration, the accuracy and stability of registration are improved, and high-precision three-dimensional image registration is realized. Meanwhile, through two-dimensional cell segmentation and three-dimensional cell matching, accurate alignment and segmentation of cell images are automatically completed, the requirement for manual intervention is reduced, and segmentation efficiency and consistency are improved.

Description

Technical Field

[0001] The present invention relates to the field of biomedical image processing, and in particular to a system, method and application for processing multiple rounds of three-dimensional biological tissue images. Background Art

[0002] In the biomedical field, in order to obtain more biological information of interest, it is necessary to perform multiple rounds of imaging on the same sample. However, due to the movement and deformation of the sample during multiple rounds of imaging, as well as the influence of different imaging conditions, the images obtained often have spatial deviations, which poses great difficulties for subsequent data analysis and information extraction. Therefore, the precise registration of multiple rounds of three-dimensional images has become an important issue in biomedical image processing. Traditional image registration methods are mainly based on feature point matching and optimization algorithms, but due to the complexity and variability of cell images, these methods alone often show limitations when processing three-dimensional cell images.

[0003] In addition, 3D cell segmentation is another key technology in biomedical image processing. Accurate cell segmentation can provide a basis for further analysis of cell characteristics. However, most existing 3D cell segmentation methods rely on algorithms based on deep learning. Although these algorithms have improved the accuracy of segmentation to a certain extent, there are still some problems and challenges: difficulty in data acquisition and annotation, high consumption of computing resources, and diversity of cell morphology, resulting in low accuracy and completeness of segmentation results.

[0004] In summary, there is an urgent need for an innovative system and method that can simultaneously solve the problems of 3D image registration and 3D cell segmentation in multiple rounds of imaging, so as to improve the accuracy and efficiency of multi-round imaging data processing. Summary of the invention

[0005] In view of the above-mentioned defects in the prior art, the present invention provides an innovative system and method that can simultaneously solve the problems of three-dimensional image registration and three-dimensional cell segmentation in multiple rounds of imaging.

[0006] To this end, the present invention provides a method for processing multiple rounds of three-dimensional biological tissue images, including multiple rounds of three-dimensional image registration and three-dimensional cell segmentation.

[0007] Among them, the multi-round three-dimensional image registration includes the following steps:

[0008] (1) Obtaining the original image;

[0009] (2) Obtain the central layer of the reference image;

[0010] (3) Obtain the central layer of the floating image;

[0011] (4) Perform layer registration on images from different rounds in sequence;

[0012] (5) Roughly align the images of the same layer;

[0013] (6) performing fine registration on the roughly registered images to obtain multiple rounds of registered three-dimensional images;

[0014] 3D cell segmentation includes the following steps:

[0015] (1) Obtaining original cell images;

[0016] (2) Filtering and denoising the original image;

[0017] (3) Segment each layer of 2D cells and obtain masks;

[0018] (4) Cell masks of different layers are matched and merged to obtain a three-dimensional cell segmentation image.

[0019] In a preferred embodiment of the present invention, multiple rounds of three-dimensional image registration specifically include:

[0020] (1) Obtaining the original image;

[0021] (2) Obtain the reference round image and calculate the average gray value of each layer, where the layer with the largest average value is the central layer;

[0022] (3) The mean square error (MSE) of each layer image of different floating rounds and the central layer image of the reference round image is traversed to obtain the layer with the smallest value as the middle layer, and the sum of the differences between the adjacent layers including it and the central layer of the reference round is calculated after normalization. The layer with the smallest sum of differences is the central layer;

[0023] (4) Perform layer registration on images from different rounds in sequence;

[0024] (5) obtaining the three-dimensional image after registration in step (4), performing a rigid transformation based on gray values ​​on the floating image and the reference image belonging to the same layer to obtain a roughly registered image;

[0025] (6) Using a SIFT-based algorithm, feature point detection is performed on the image after the rough registration in step (5), the SIFT descriptor of each feature point is calculated, feature point matching is performed between the floating image and the reference image, and the affine transformation parameters from the floating image to the reference image are estimated. The estimated geometric transformation parameters are applied to finely align the image after the rough registration, and multiple rounds of three-dimensional images after z-axis and xy-axis registration are obtained.

[0026] In a further preferred embodiment of the present invention, three-dimensional cell segmentation specifically comprises:

[0027] (1) Obtaining original cell images;

[0028] (2) Perform Gaussian filtering to denoise the two-dimensional cell images of each layer;

[0029] (3) Cell segmentation is performed on each layer of the image based on the grayscale threshold of the layer, the holes in the cell nucleus are filled based on morphological operations, and the connected cells are separated by the watershed algorithm to segment the independent cell regions and obtain the cell mask;

[0030] (4) Matching and merging the cell masks on the two-dimensional slices obtained in step (3) into three-dimensional cells, generating a three-dimensional cell segmentation result, and obtaining a three-dimensional cell segmentation image.

[0031] In a more preferred embodiment of the present invention, step (4) is specifically:

[0032] The convex hull of the segmented cell area on each slice is calculated, and the intersection-and-union ratio matrix of the cell convex hulls between adjacent slices is calculated. The weights of the grayscale intensity and area characteristics of the cells are comprehensively considered, and the Hungarian algorithm is used to find the best cell pairing between adjacent slices. Based on the pairing results of the Hungarian algorithm, the cell areas on adjacent slices are matched to generate a three-dimensional cell segmentation result, and the masks belonging to the same cell are marked with the same color to achieve three-dimensional visualization of cell segmentation and obtain a three-dimensional cell segmentation image.

[0033] In the most preferred embodiment of the present invention, the specific operation of step (4) is as shown in formulas (1) and (2):

[0034]

[0035] Among them, Am and Bn are the cell convex hulls, where m and n represent the numbers of the two cells respectively.

[0036]

[0037] Comprehensive scoring matrix S mn It combines IoU (intersection over union), grayscale intensity and area features, where I(A m ) and I(B n ) represent cell A m With B n The average gray intensity, Area(A m ) and Area(B n ) represent cell A m With B n The area, w IoU 、w I and w A is the weight coefficient.

[0038] Another aspect of the present invention provides a system for processing multiple three-dimensional biological tissue images, including a multiple three-dimensional image registration module and a three-dimensional cell segmentation module.

[0039] Among them, the multi-round 3D image registration module includes the following sub-modules:

[0040] (1) an original image acquisition module, used to acquire the original image;

[0041] (2) a reference image center layer acquisition module, used to acquire the center layer of the reference image;

[0042] (3) A floating image center layer acquisition module, used to acquire the center layer of the floating image;

[0043] (4) a layer registration module, used to sequentially register the layers of images from different rounds;

[0044] (5) a coarse registration module, used to coarsely register images of the same layer;

[0045] (6) a fine registration and multi-round three-dimensional image acquisition module, which is used to perform fine registration on the roughly registered image and obtain the registered multi-round three-dimensional image;

[0046] The 3D cell segmentation module includes the following submodules:

[0047] (1) a raw cell image acquisition module, used to acquire raw cell images;

[0048] (2) an original image filtering and denoising module, used to filter and denoise the original image;

[0049] (3) Segmentation and mask acquisition module, used to segment each layer of 2D cells and obtain masks;

[0050] (4) A mask matching and merging and three-dimensional cell segmentation image acquisition module is used to match and merge cell masks of different layers to obtain a three-dimensional cell segmentation image.

[0051] In a preferred embodiment of the present invention, the multi-round three-dimensional image registration module specifically includes:

[0052] (1) an original image acquisition module, used to acquire the original image;

[0053] (2) A reference image center layer acquisition module, which is used to obtain the reference round image and calculate the average gray value of each layer, where the layer with the largest average value is the center layer;

[0054] (3) A floating image center layer acquisition module is used to traverse each layer image of different floating rounds and the center layer image of the reference round image to calculate the mean square error (MSE), obtain the layer with the smallest value as the middle layer, and calculate the sum of the differences between the adjacent layers including it and the center layer of the reference round after normalization. The layer with the smallest sum of differences is the center layer;

[0055] (4) a layer registration module, used to sequentially register the layers of images from different rounds;

[0056] (5) a coarse registration module, which is used to obtain the registered three-dimensional image, and to perform a rigid transformation based on gray values ​​on the floating image and the reference image belonging to the same layer to obtain the coarsely registered image;

[0057] (6) A fine registration and multi-round three-dimensional image acquisition module, which is used to detect feature points of the image after coarse registration using a SIFT-based algorithm, calculate the SIFT descriptor of each feature point, match feature points between the floating image and the reference image, estimate the affine transformation parameters from the floating image to the reference image, apply the estimated geometric transformation parameters, perform fine registration on the image after coarse registration, and obtain multi-round three-dimensional images after z-axis and xy-axis registration.

[0058] In a more preferred embodiment of the present invention, the three-dimensional cell segmentation module specifically includes:

[0059] (1) a raw cell image acquisition module, used to acquire raw cell images;

[0060] (2) Original image filtering and denoising module, used to perform Gaussian filtering and denoising on the two-dimensional cell images of each layer;

[0061] (3) Segmentation and mask acquisition module, which is used to perform cell segmentation on each layer of the image based on the grayscale threshold of the layer, fill the holes in the cell nucleus based on morphological operations, separate the connected cells using the watershed algorithm, segment the independent cell area and obtain the cell mask;

[0062] (4) A mask matching and merging and three-dimensional cell segmentation image acquisition module, which is used to match and merge the cell masks on the two-dimensional slices obtained by module (3) into three-dimensional cells, generate three-dimensional cell segmentation results, and obtain a three-dimensional cell segmentation image.

[0063] Another aspect of the present invention provides application of the method or system described in the present invention in processing multiple rounds of three-dimensional biological tissue images.

[0064] In a preferred embodiment of the present invention, the application is an application in fluorescent in situ hybridization imaging of Caenorhabditis elegans or an application in spatial translationomics imaging of mouse coronal brain slices.

[0065] The present invention achieves the following beneficial effects by adopting the above technical solution:

[0066] 1. The present invention uses multi-step registration, including gray value mean square error calculation and SIFT feature point matching, to fully utilize the information of gray values ​​and feature points, improve the accuracy and stability of registration, achieve high-precision three-dimensional image registration, and make the registration process more robust and able to cope with image rotation and scaling changes. In addition, the method can automatically complete the registration process of multiple rounds of images, reduce the need for manual intervention, and improve efficiency and consistency.

[0067] 2. The present invention can automatically complete the precise alignment and segmentation of cell images through two-dimensional cell segmentation and three-dimensional cell matching, reducing the need for manual intervention and improving segmentation efficiency and consistency. Especially for dense cell areas, the present invention can achieve more accurate cell segmentation and provide clearer and more detailed cell contour information, which is suitable for various biomedical research and clinical applications, and improves the quality and effect of cell analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 The present invention is a flowchart of the steps of processing multiple rounds of three-dimensional biological tissue images.

[0069] Figure 2 This is a system framework diagram for processing multiple rounds of three-dimensional biological tissue images in the present invention.

[0070] Figure 3 The original image of OFP mRNA fluorescence in situ hybridization imaging of Caenorhabditis elegans and the image result after multiple rounds of three-dimensional registration using the present invention are shown.

[0071] A: reference image;

[0072] B: floating image;

[0073] C: The middle layer after fine registration;

[0074] D: center layer of the reference image;

[0075] E: The center layer of the floating image;

[0076] F: reference image after fine registration;

[0077] G: Image after fine registration.

[0078] Figure 4 The original image of the cell nucleus of Caenorhabditis elegans stained with DAPI and the result of three-dimensional cell segmentation using the present invention are shown.

[0079] A: DAPI staining results of cell nuclei;

[0080] B: Image after filtering and denoising;

[0081] C: Cell segmentation and mask results;

[0082] D: Original image of three-dimensional cells;

[0083] E: 3D cell segmentation image.

[0084] Figure 5 The original image of spatial translationomics imaging of mouse coronal brain slices and the resulting image after multiple rounds of three-dimensional alignment using the present invention are shown.

[0085] A: reference image;

[0086] B: floating image;

[0087] C: The middle layer after fine registration;

[0088] D: center layer of the reference image;

[0089] E: The center layer of the floating image;

[0090] F: reference image after fine registration;

[0091] G: Image after fine registration.

[0092] Figure 6 The original image of spatial translationomics imaging of mouse coronal brain slices and the result of three-dimensional cell segmentation using the present invention are shown.

[0093] A: DAPI staining results of cell nuclei;

[0094] B: Image after filtering and denoising;

[0095] C: Cell segmentation and mask results;

[0096] D: Original image of three-dimensional cells;

[0097] E: 3D cell segmentation image. DETAILED DESCRIPTION

[0098] The present application is further described in detail below through the accompanying drawings and embodiments. Through these descriptions, the characteristics and advantages of the present application will become clearer and more specific.

[0099] Example 1: Application of the system and method for processing multiple rounds of three-dimensional biological tissue images of the present invention in fluorescent in situ hybridization imaging of Caenorhabditis elegans

[0100] The method and system for processing multiple rounds of three-dimensional biological tissue images in this embodiment are respectively as shown in the attached drawings. Figure 1 and attached Figure 2As shown, the details are as follows.

[0101] (I) Multi-round 3D image registration technology

[0102] 1. Get the original image

[0103] The imaging of Caenorhabditis elegans based on fluorescence in situ hybridization imaging was obtained using a scanning confocal microscope; the z-stack z-axis step length was 0.9 μm, and the obtained image was a grayscale image with pseudo-color, and the 405 channel imaging was a DAPI staining image, and the 488 channel was a fluorescence imaging of OFP mRNA. Different rounds of OFP mRNA fluorescence imaging were used for multiple rounds of three-dimensional registration, and DAPI staining images were used for cell segmentation. The specific original images obtained are shown in the attached figure. Figure 3 A and 3B, wherein the Figure 3 A is the reference image, Figure 3 B is a floating image.

[0104] 2. Get the center layer of the reference image

[0105] Calculate the grayscale average of each layer of the reference image. The layer with the largest grayscale average is the central layer of the reference image. Figure 3 As shown in D.

[0106] 3. Get the center layer of the floating image

[0107] The mean square error (MSE) of each layer of the floating image and the central layer of the reference image is calculated, and the layer with the smallest value is taken as the middle layer. The sum of the differences between the normalized three adjacent layers including the middle layer and the normalized central layer of the reference image is calculated, and the layer with the smallest sum of differences is taken as the central layer. Figure 3 E. The normalized difference between adjacent layers is calculated to further determine the most matching central layer among similar layers.

[0108] 4. Align the layers of images from different rounds in sequence

[0109] After pairing the reference image with the central layer of the floating image, the adjacent layers are paired in turn, and the images of the redundant unpaired layers are discarded to obtain round images with the same dimension.

[0110] 5. Roughly align images on the same layer

[0111] The floating image and the reference image belonging to the same layer are rigidly transformed based on the grayscale value to obtain a roughly registered image. Specifically, the "imregister" function based on Matlab and the "translation" transformation form are used for registration to obtain a roughly registered image. By transforming each layer of the image in turn, the movement of the C. elegans samples in different rounds and the movement in the field of view can be eliminated.

[0112] 6. Finely align the coarsely aligned images

[0113] The coarsely registered floating image is finely registered based on SIFT feature extraction, and the affine transformation parameters from the floating image to the reference image are estimated based on the matched feature point pairs. The RANSAC algorithm is usually used to ensure the robustness of the transformation and remove false matches.

[0114] Here, problems such as local deformation caused by osmotic pressure between the C. elegans sample and the solution can be eliminated.

[0115] Obtain multiple rounds of 3D images after registration. The finely registered images are shown in the attached Figure 3 C. Attachment Figure 3 F and attached Figure 3 G. Figure 3 C is the middle layer after fine registration. Figure 3 F is the reference image, Figure 3 G is the image after fine registration.

[0116] (II) 3D cell segmentation technology

[0117] 1. Obtain original cell images

[0118] Get the DAPI staining result of Caenorhabditis elegans nucleus under channel 405, as shown in the attached figure. Figure 4 As shown in A. The imaging device is a scanning confocal microscope, wherein the z-axis step length of the z-stack is 0.9 μm, the z-stack has 59 layers, of which 40 layers contain effective cell nucleus layers.

[0119] 2. Original image filtering and denoising

[0120] Gaussian filtering is performed on each layer to remove noise from the stained image. This can be achieved based on the "imgaussfilt" function in Matlab, and the sigma value and filter size are set as small as possible to retain the image details to the greatest extent for subsequent cell segmentation. The image after filtering and denoising is shown in the attached figure. Figure 4 As shown in B.

[0121] 3. Segment each layer of 2D cells and obtain masks

[0122] Grayscale thresholds were applied to each layer of the image for cell segmentation, and the holes in the cell nuclei were filled based on morphological operations. The watershed algorithm was applied to separate the connected cells to segment the independent cell regions and obtain the cell masks. The results are shown in the attached figure. Figure 4 As shown in C.

[0123] 4. Cell mask matching and merging of different layers

[0124] The cell masks on the obtained two-dimensional slices are matched and merged into three-dimensional cells. Specifically, the convex hull of the segmented cell mask on each slice is calculated, and the intersection over union (IoU) matrix of the cell convex hulls between adjacent slices is calculated. The weights of the grayscale intensity and area features of the cells are comprehensively considered, and the Hungarian algorithm is used to find the best cell pairing between adjacent slices. Based on the pairing results of the Hungarian algorithm, the cell areas on adjacent slices are matched and assigned the same grayscale value labels to generate a three-dimensional cell segmentation result.

[0125] Get the 3D cell segmentation image. The result is as shown in the attached figure. Figure 4 D and 4E.

[0126] Example 2: Application of the system and method for processing multiple rounds of three-dimensional biological tissue images in spatial translational omics imaging of coronal brain slices of mice

[0127] The method and system for processing multiple rounds of three-dimensional biological tissue images in this embodiment are respectively as shown in the attached drawings. Figure 1 and attached Figure 2 As shown, the details are as follows.

[0128] (I) Multi-round 3D image registration technology

[0129] 1. Get the original image

[0130] Coronal brain slice images of mice were obtained for spatial translationomics imaging. The imaging device was a scanning confocal microscope. The z-axis step length of the z-stack was 0.1 μm. The images obtained were grayscale images with pseudo colors. The 405 channel images were DAPI-stained images, and the 488 channel images were fluorescence images of the MALAT-1 mRNA being translated. Different rounds of MALAT-1 mRNA fluorescence imaging were used for multiple rounds of three-dimensional registration, and DAPI-stained images were used for cell segmentation. The specific original images obtained are shown in the attached figure. Figure 5 A and 5B, wherein the Figure 5 A is the reference image, Figure 5 B is a floating image.

[0131] 2. Get the center layer of the reference image

[0132] Calculate the grayscale average of each layer of the reference image. The layer with the largest grayscale average is the central layer of the reference image. Figure 5 As shown in D.

[0133] 3. Get the center layer of the floating image

[0134] The mean square error (MSE) of each layer of the floating image and the central layer of the reference image is calculated, and the layer with the smallest value is taken as the middle layer. The sum of the differences between the normalized three adjacent layers including the middle layer and the normalized central layer of the reference image is calculated, and the layer with the smallest sum of differences is taken as the central layer. Figure 5 E. The normalized difference between adjacent layers is calculated to further determine the most matching central layer among similar layers.

[0135] 4. Align the layers of images from different rounds in sequence

[0136] After pairing the reference image with the center layer of the floating image, the adjacent layers are paired in turn, and the redundant images without paired layers are discarded to obtain round images with the same dimension.

[0137] 5. Roughly align images on the same layer

[0138] The floating image and the reference image belonging to the same layer are rigidly transformed based on the gray value to obtain a roughly registered image. Specifically, the "imregister" function based on Matlab and the "translation" transformation form are used for registration to obtain a roughly registered image. By transforming each layer of the image in turn, the movement of the mouse coronal brain slice samples in different rounds and the movement in the field of view space can be eliminated.

[0139] 6. Finely align the coarsely aligned images

[0140] The coarsely registered floating image is finely registered based on SIFT feature extraction, and the affine transformation parameters from the floating image to the reference image are estimated based on the matched feature point pairs. The RANSAC algorithm is usually used to ensure the robustness of the transformation and remove false matches.

[0141] This can eliminate problems such as local deformation caused by osmotic pressure between the mouse coronal brain slice sample and the solution.

[0142] Obtain multiple rounds of 3D images after registration. The finely registered images are shown in the attached Figure 5 C. Attachment Figure 5 F and attached Figure 5 G. Figure 5 C is the middle layer after fine registration. Figure 5F is the reference image, Figure 5 G is the image after fine registration.

[0143] (II) 3D cell segmentation technology

[0144] 1. Obtain original cell images

[0145] Get the DAPI staining results of the mouse coronal brain slice nucleus under channel 405, as shown in the attached figure. Figure 6 As shown in A. The imaging device is a scanning confocal microscope, wherein the z-axis step length of the z-stack is 0.1 μm, the z-stack has 17 layers, and the layers containing effective cell nuclei are 17 layers.

[0146] 2. Original image filtering and denoising

[0147] Gaussian filtering is performed on each layer to remove noise from the stained image. This can be achieved based on the "imgaussfilt" function in Matlab, and the sigma value and filter size are set as small as possible to retain the image details to the greatest extent for subsequent cell segmentation. The image after filtering and denoising is shown in the attached figure. Figure 6 As shown in B.

[0148] 3. Segment each layer of 2D cells and obtain masks

[0149] Grayscale thresholds were applied to each layer of the image for cell segmentation, and the holes in the cell nuclei were filled based on morphological operations. The watershed algorithm was applied to separate the connected cells to segment the independent cell regions and obtain the cell masks. The results are shown in the attached figure. Figure 6 As shown in C.

[0150] 4. Cell mask matching and merging of different layers

[0151] The cell masks on the obtained two-dimensional slices are matched and merged into three-dimensional cells. Specifically, the convex hull of the segmented cell mask on each slice is calculated, and the intersection over union (IoU) matrix of the cell convex hulls between adjacent slices is calculated. The weights of the grayscale intensity and area features of the cells are comprehensively considered, and the Hungarian algorithm is used to find the best cell pairing between adjacent slices. Based on the pairing results of the Hungarian algorithm, the cell areas on adjacent slices are matched and assigned the same grayscale value labels to generate a three-dimensional cell segmentation result.

[0152] Get the 3D cell segmentation image. The result is as shown in the attached figure. Figure 6 D and 6E.

[0153] The present application has been described above in conjunction with preferred embodiments, but these embodiments are only exemplary and serve only as an illustration. On this basis, various replacements and improvements may be made to the present application, all of which fall within the scope of protection of the present application.

Claims

1. A method for processing multiple rounds of three-dimensional biological tissue images, including multiple rounds of three-dimensional image registration and three-dimensional cell segmentation, in, The multi-round 3D image registration includes the following steps: (1) Obtaining the original image; (2) Obtain the central layer of the reference image; (3) Obtain the central layer of the floating image; (4) Perform layer registration on images from different rounds in sequence; (5) Roughly align the images of the same layer; (6) performing fine registration on the roughly registered images to obtain multiple rounds of registered three-dimensional images; 3D cell segmentation includes the following steps: (1) Obtaining original cell images; (2) Filtering and denoising the original image; (3) Segment each layer of 2D cells and obtain masks; (4) Cell masks of different layers are matched and merged to obtain a three-dimensional cell segmentation image.

2. The method according to claim 1, wherein: The multiple rounds of 3D image registration specifically include: (1) Obtaining the original image; (2) Obtain the reference round image and calculate the average gray value of each layer, where the layer with the largest average value is the central layer; (3) The mean square error (MSE) of each layer image of different floating rounds and the central layer image of the reference round image is traversed to obtain the layer with the smallest value as the middle layer, and the sum of the differences between the adjacent layers including it and the central layer of the reference round is calculated after normalization. The layer with the smallest sum of differences is the central layer; (4) Perform layer registration on images from different rounds in sequence; (5) obtaining the three-dimensional image after registration in step (4), performing a rigid transformation based on gray values ​​on the floating image and the reference image belonging to the same layer to obtain a roughly registered image; (6) Using a SIFT-based algorithm, feature point detection is performed on the image after the rough registration in step (5), the SIFT descriptor of each feature point is calculated, feature point matching is performed between the floating image and the reference image, and the affine transformation parameters from the floating image to the reference image are estimated. The estimated geometric transformation parameters are applied to finely align the image after the rough registration, and multiple rounds of three-dimensional images after z-axis and xy-axis registration are obtained.

3. The method according to claim 1 or 2, wherein: 3D cell segmentation specifically includes: (1) Obtaining original cell images; (2) Perform Gaussian filtering to denoise the two-dimensional cell images of each layer; (3) Cell segmentation is performed on each layer of the image based on the grayscale threshold of the layer, the holes in the cell nucleus are filled based on morphological operations, and the connected cells are separated by the watershed algorithm to segment the independent cell regions and obtain the cell mask; (4) Matching and merging the cell masks on the two-dimensional slices obtained in step (3) into three-dimensional cells, generating a three-dimensional cell segmentation result, and obtaining a three-dimensional cell segmentation image.

4. The method according to claim 3, wherein step (4) specifically comprises: The convex hull of the segmented cell area on each slice is calculated, and the intersection-and-union ratio matrix of the cell convex hulls between adjacent slices is calculated. The weights of the grayscale intensity and area characteristics of the cells are comprehensively considered, and the Hungarian algorithm is used to find the best cell pairing between adjacent slices. Based on the pairing results of the Hungarian algorithm, the cell areas on adjacent slices are matched to generate a three-dimensional cell segmentation result, and the masks belonging to the same cell are marked with the same color to achieve three-dimensional visualization of cell segmentation and obtain a three-dimensional cell segmentation image.

5. The method according to claim 4, wherein the specific operation of step (4) is as shown in formulas (1) and (2): in, Am and Bn are the cell convex hulls, where m and n represent the numbers of the two cells respectively. Comprehensive scoring matrix S mn It combines IoU (intersection over union), grayscale intensity and area features, where I(A m ) and I(B n ) represent cell A m With B n The average gray intensity, Area(A m ) and Area(B n ) represent cell A m With B n The area, w IoU 、w I and w A is the weight coefficient.

6. A system for processing multiple three-dimensional biological tissue images, comprising a multiple three-dimensional image registration module and a three-dimensional cell segmentation module, in, The multi-round 3D image registration module includes the following sub-modules: (1) an original image acquisition module, used to acquire the original image; (2) a reference image center layer acquisition module, used to acquire the center layer of the reference image; (3) A floating image center layer acquisition module, used to acquire the center layer of the floating image; (4) a layer registration module, used to sequentially register the layers of images from different rounds; (5) a coarse registration module, used to coarsely register images of the same layer; (6) a fine registration and multi-round three-dimensional image acquisition module, which is used to perform fine registration on the roughly registered image and obtain the registered multi-round three-dimensional image; The 3D cell segmentation module includes the following submodules: (1) a raw cell image acquisition module, used to acquire raw cell images; (2) an original image filtering and denoising module, used to filter and denoise the original image; (3) Segmentation and mask acquisition module, used to segment each layer of 2D cells and obtain masks; (4) A mask matching and merging and three-dimensional cell segmentation image acquisition module is used to match and merge cell masks of different layers to obtain a three-dimensional cell segmentation image.

7. The system according to claim 6, wherein the multi-round three-dimensional image registration module specifically comprises: (1) an original image acquisition module, used to acquire the original image; (2) A reference image center layer acquisition module, which is used to obtain the reference round image and calculate the average gray value of each layer, where the layer with the largest average value is the center layer; (3) A floating image center layer acquisition module is used to traverse each layer image of different floating rounds and the center layer image of the reference round image to calculate the mean square error (MSE), obtain the layer with the smallest value as the middle layer, and calculate the sum of the differences between the adjacent layers including it and the center layer of the reference round after normalization. The layer with the smallest sum of differences is the center layer; (4) a layer registration module, used to sequentially register the layers of images from different rounds; (5) a coarse registration module, which is used to obtain the registered three-dimensional image, and to perform a rigid transformation based on gray values ​​on the floating image and the reference image belonging to the same layer to obtain the coarsely registered image; (6) A fine registration and multi-round three-dimensional image acquisition module, which is used to detect feature points of the image after coarse registration using a SIFT-based algorithm, calculate the SIFT descriptor of each feature point, match feature points between the floating image and the reference image, estimate the affine transformation parameters from the floating image to the reference image, apply the estimated geometric transformation parameters, perform fine registration on the image after coarse registration, and obtain multi-round three-dimensional images after z-axis and xy-axis registration.

8. The system according to claim 6 or 7, wherein the three-dimensional cell segmentation module specifically comprises: (1) a raw cell image acquisition module, used to acquire raw cell images; (2) Original image filtering and denoising module, used to perform Gaussian filtering and denoising on the two-dimensional cell images of each layer; (3) Segmentation and mask acquisition module, which is used to perform cell segmentation on each layer of the image based on the grayscale threshold of the layer, fill the holes in the cell nucleus based on morphological operations, separate the connected cells using the watershed algorithm, segment the independent cell area and obtain the cell mask; (4) A mask matching and merging and three-dimensional cell segmentation image acquisition module, which is used to match and merge the cell masks on the two-dimensional slices obtained by module (3) into three-dimensional cells, generate three-dimensional cell segmentation results, and obtain a three-dimensional cell segmentation image.

9. Use of the method according to any one of claims 1 to 5 or the system according to any one of claims 6 to 8 in processing multiple rounds of three-dimensional biological tissue images. 10 . The use according to claim 9 , which is use in fluorescent in situ hybridization imaging of Caenorhabditis elegans or use in spatial translationomics imaging of mouse coronal brain slices.

Images