Tissue subcellular level three-dimensional reconstruction method

Abstract

The invention discloses a tissue subcellular level three-dimensional reconstruction method, and belongs to the technical field of image recognition and three-dimensional reconstruction. According to the specific scheme, the method comprises the steps of image preprocessing, image segmentation, region analysis and classification, cell tracking and screening, region extraction and marking and three-dimensional image visualization. According to the method, the three-dimensional space relation of a multi-layer cell structure and two-dimensional morphological characteristics in a slice image are combined, and accurate recognition of cells and tissues is achieved through multi-dimensional indexes; three-dimensional information obtained by high-precision continuous slicing is fully utilized, and a high-quality three-dimensional reconstruction result is obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition and three-dimensional reconstruction, and in particular relates to a tissue subcellular level three-dimensional reconstruction method. Background Art

[0002] 3D reconstruction of bone pathology is a key technology in orthopedic research and clinical diagnosis. By continuously sectioning tissue to obtain a series of tomographic images, the three-dimensional structure of the tissue can be reconstructed. Compared to traditional two-dimensional slice observation, 3D reconstruction of tissue structure can provide more intuitive and richer information, greatly assisting manual observation, analysis, and clinical diagnosis.

[0003] Common 3D reconstruction methods convert registered images into grayscale images and stack them into voxel data. Triangular mesh surfaces are then extracted using isosurface extraction algorithms, such as Marching Cubes. Alternatively, image segmentation algorithms are used to perform region division, which is then stacked to create a 3D structure. The effectiveness of traditional tissue 3D reconstruction is limited by the fact that the slice thickness approaches the cell size. This often makes it difficult to continuously track the same cellular structure across multiple images. Image registration can only be performed based on more macroscopic structural features, without guaranteeing the accuracy of cell position and the integrity of morphology. With advancements in tissue sectioning technology, the minimum tissue section thickness can now reach the subcellular level, providing hardware support for tracking the same cell across multiple images. However, due to image quality limitations, region division in traditional 3D reconstruction methods can be inaccurate, potentially resulting in erroneous results and compromising 3D reconstruction quality. Therefore, a subcellular 3D reconstruction method based on multi-layer data association, analyzing image segmentation results, and assisting in filtering out erroneous region division results, ultimately achieving precise subcellular spatial observation of tissues, is currently lacking. Summary of the Invention

[0004] The primary problem with this method is that the 3D reconstruction based on serial slice images fails to examine the 3D spatial connections between multilayered cellular structures at the subcellular scale, making it impossible to accurately identify and locate the same cellular structure. A secondary problem is that the 3D reconstruction using region segmentation methods lacks analysis of the 2D morphological features of the region segmentation results, such as area and circularity calculations.

[0005] The present invention provides a method for 3D reconstruction of tissue at the subcellular level. It combines the 3D spatial connection of multi-layered cell structures with the 2D morphological features in slice images, achieves accurate identification of cells and tissues through multi-dimensional indicators, and fully utilizes the 3D information obtained from high-precision continuous slices to obtain high-quality 3D reconstruction results.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for three-dimensional reconstruction of tissue subcellular level, comprising the following steps:

[0008] Step 1: Image Preprocessing

[0009] Preprocessing of original tissue (e.g., bone tissue) slice images to improve regional segmentation accuracy. Specifically, the images captured by the microscope camera are converted to grayscale images. These grayscale images are enhanced using contrast-constrained adaptive histogram equalization (LCAE). This method adjusts the image contrast by region to improve overall image legibility.

[0010] Step 2: Image Segmentation

[0011] The K-means algorithm and watershed algorithm are used to divide and segment the preprocessed image respectively to extract the area where the cells are located.

[0012] First, the K-means algorithm is used to divide the image into two parts, and the result of the two-part division is converted into a binary image. Morphological dilation, erosion and image difference operations are performed to obtain the image foreground, background and the area to be determined. Based on the above three types of areas, initial labels are generated for watershed segmentation. The result of the watershed algorithm is a label image, and different areas are assigned independent pixel value labels. According to different pixel values, the pixel set of each area is obtained. i}.

[0013] Step 3: Regional analysis and classification

[0014] The morphological characteristic indicators of each area were calculated, different areas were distinguished according to these indicators, and the cells were preliminarily screened.

[0015] Calculate the area, centroid, and circularity of each region:

[0016] Area A: A = n

[0017] Centroid C:

[0018] Roundness e:

[0019] Where n is the number of pixels in the area, (x i ,y i ) is the pixel coordinate, and P is the region perimeter. The region perimeter can be obtained by tracking the region boundary.

[0020] Set thresholds for the above indicators and i} to filter and obtain a new set {R' j Based on the size, the matrix area can be further distinguished from the cell area.

[0021] Step 4: Cell tracking screening

[0022] For a cell region in a certain cross-sectional image, the cell is tracked in several layers of images before and after, and the misidentified isolated areas are filtered out based on the tracking results to achieve accurate cell identification.

[0023] By calculating the area R of the current layer i With the region R in another layer j The intersection over union (IoU) is used for cell tracking.

[0024] Intersection over Union (IoU):

[0025] Where A(R) represents the area of the region.

[0026] Each layer's regions are compared with those in the preceding and following layers, and the maximum Intersection over Union (IoU) between the regions is recorded. Weights are set based on the inter-layer distance, with larger inter-layer distances resulting in smaller weights. The weighted IoU sum is used as the basis for selecting cell regions. If a region is not effectively tracked in the preceding and following layers (i.e., the sum of the IoU values is small or close to 0), it is considered an incorrect segmentation result and is removed.

[0027] Step 5: Region extraction and labeling

[0028] Based on the results of region segmentation and screening, a mask is generated to extract the corresponding region in the original image. Different regions (such as bone matrix and bone cells) are labeled with different labels to facilitate subsequent analysis.

[0029] Step 6: 3D Image Visualization

[0030] Multiple annotated serial slice images are stacked in spatial order to reconstruct 3D volume data, which is then imported into a 3D image viewer. The XYZ coordinate scale is adjusted based on the correspondence between pixels and physical dimensions, and different structures can be displayed based on region labels, such as only bone cells.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] This method performs regional segmentation on serial slice images of tissue samples, examining the 2D morphological characteristics of each region and the 3D spatial connections between multi-layer regions to achieve precise tissue cell segmentation. Using a 1600*1600 grayscale image as input, regional segmentation can be completed within 1 second, with an accuracy rate exceeding 80%. In multi-layer images, continuous changes within the same cell region are clearly visible. Based on this, 3D structural reconstruction is performed, resulting in clear cellular structures and complete morphology. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is the overall algorithm flow chart of the present invention;

[0034] Figure 2 This is the result diagram of cell area division;

[0035] Figure 3 is the 3D reconstruction result. DETAILED DESCRIPTION

[0036] The technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings and embodiments. Obviously, the described embodiments are only part of the embodiments of the invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0037] Example 1:

[0038] Take the mouse femur sample as an example. Cut continuously at 1 micron intervals to obtain 20 layers of cross-sectional images. The specific steps are as follows Figure 1 As shown,

[0039] In step 1, a 10*10 window is used and the contrast is limited to 3.0. CLAHE is performed on the image with this parameter setting, which can significantly improve the image contrast without excessively enhancing some interference information.

[0040] In step 2, the image is segmented into three main regions: bone matrix (large area with high grayscale), osteocyte lacunae (small area with low grayscale), and cavity (large area with low grayscale). Bone matrix and osteocyte lacunae are the regions of interest, so they are selected based on grayscale and area.

[0041] In step 3, some misidentified areas can be screened based on morphological features, such as extremely small noise points (area screening) and narrow knife marks (roundness screening). The remaining areas with similar size and morphology to cells are further distinguished in step 4.

[0042] In step 4, the first three layers and the last three layers of a given image are analyzed, with weights ranging from 0.5, 0.3, and 0.2, respectively. The sum of the larger IoU between the two layers with the same distance to the current layer is calculated. It is clear that the maximum value of this sum is 1, corresponding to the case where the layers have completely identical regions. In this case, since most of the mis-segmented areas are isolated, a smaller threshold (less than 0.1) can effectively filter out these areas.

[0043] Steps 5 and 6 are consistent with the content of the invention. The results of cell area division and final three-dimensional reconstruction are as follows Figure 2 and Figure 3 shown.

[0044] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A method for three-dimensional reconstruction of tissue subcellular level, characterized in that: The following steps are involved: Step 1: Image preprocessing: convert the tissue section image into a grayscale image, enhance the grayscale image, and adjust the image contrast by region; Step 2: Image segmentation: Use K-means algorithm and watershed algorithm to divide and segment the preprocessed image respectively; Step 3: Regional analysis and classification: Calculate the morphological characteristics of each region, distinguish different regions based on these indicators, and conduct preliminary screening of cells; Step 4: Cell tracking screening: For a cell region in a certain cross-sectional image, track the cell in several layers of images before and after, and filter out isolated regions that were misidentified based on the tracking results to achieve accurate cell identification; Step 5: Region extraction and labeling: Generate a mask based on the region segmentation and screening results, extract the corresponding region in the original image, and use different labels to mark different types of regions; Step 6: 3D image visualization: stack the annotated continuous slice images in spatial order, reconstruct them into 3D volume data, and import them into a 3D image viewer.

2. The method for three-dimensional reconstruction of tissue subcellular level according to claim 1, characterized in that: In step 1, the grayscale image is enhanced using the contrast-constrained adaptive histogram equalization method.

3. The method for three-dimensional reconstruction of tissue subcellular level according to claim 1, characterized in that: The specific steps of step 2 are as follows: first, use the K-means algorithm to divide the image into two parts, convert the result of the division into a binary image, perform morphological dilation, erosion and image difference operations to obtain the image foreground, background and the area to be determined; generate initial labels based on the above three types of areas and perform watershed segmentation; the result of the watershed algorithm is a labeled image, and different areas are assigned independent pixel value labels; According to different pixel values, the pixel set of each area is obtained {R i }.

4. The method for three-dimensional reconstruction of tissue subcellular level according to claim 1, characterized in that: In step 3, the morphological characteristic indicators include area, centroid and roundness, and the calculation formula is as follows: Area A: A = n Centroid C: Roundness e: Where n is the number of pixels in the area, (x i ,y i ) are pixel coordinates, and P is the perimeter of the region.

5. The method for three-dimensional reconstruction of tissue subcellular level according to claim 1, characterized in that: In step 4, by calculating the area R of the current layer i With the region R in another layer j The intersection over union (IoU) is used for cell tracking. Intersection over Union (IoU): Where A(R) represents the area of the region. The region of each layer is compared with the regions of several previous and next layers, and the maximum intersection-over-union ratio between the regions is recorded. The weight is set according to the inter-layer distance. The larger the inter-layer distance, the smaller the weight. The sum of the weighted intersection-over-union ratios is used as the basis for cell region screening.

6. The method for three-dimensional reconstruction of tissue subcellular level according to claim 1, characterized in that: In step 6, the XYZ coordinate ratio is adjusted according to the correspondence between pixels and physical dimensions, and different structures are displayed according to the region labels.

Images