A method and system for estimating cell occlusion relationships based on pixel-level focus evaluation curves

Abstract

This invention discloses a method and system for estimating cell occlusion relationships based on pixel-level focus evaluation curves. The method includes: acquiring multi-focal-plane cell image sequences at different focus positions within the same field of view; segmenting the image sequences into cell instances to obtain a two-dimensional cell instance mask; calculating the sharpness value of each pixel within the mask on each focal plane to construct a pixel-level focus response curve; determining the focus depth range of each pixel based on the peak position of the focus response curve and a preset relative threshold; and comparing the focus depth ranges of different cell pixels in the cell overlap region to determine the vertical occlusion relationship between cells. This invention achieves automated analysis of cell vertical relationships from raw microscopic data through pixel-level depth modeling and range comparison, improving the spatial resolution and reliability of occlusion relationship determination. It also exhibits high compatibility with conventional microscopic imaging workflows and is easy to deploy and promote.

Description

Technical Field

[0001] This invention belongs to the field of cell image processing technology, specifically relating to a method and system for estimating cell occlusion relationships based on pixel-level focus evaluation curves. Background Technology

[0002] In cellular microscopy, accurately obtaining the stacking relationship of cells along the z-axis is crucial for understanding cell-cell interactions, analyzing tissue structure, and tracking cell behavior. Currently, researchers typically employ two approaches to address this problem.

[0003] The first type of method captures only one or a few focal plane images, performing cell segmentation and overlap analysis on a two-dimensional plane. Essentially, this method only discusses whether masks overlap within a single planar image, failing to truly identify the hierarchical relationships between cells at overlapping points. The second type of method captures multiple images consecutively at different focus positions, calculating sharpness indices at each layer to select the sharpest focal plane. However, these methods mostly only select the approximate optimal focal plane for each cell, or compress multi-layer information back into a single two-dimensional image using methods such as maximum projection or weighted projection. They do not convert the sharpness variations of each pixel at different focal planes into specific depth range information, nor do they systematically establish hierarchical relationships between cells at the pixel and intersection levels.

[0004] In real-world cell applications, issues such as imaging noise, uneven illumination, and differences in cell transparency are common. Existing technologies either only provide a rough overall depth impression or only improve the visual effect, failing to offer frontline researchers a set of analytical tools that can directly identify the relationships between cell layers and can be performed in batches.

[0005] In summary, existing technologies have the following shortcomings: First, existing methods generally describe the state of sharpness or blurriness on a cell-wide or whole-image basis, lacking the ability to estimate the depth range of individual pixels in local areas where overlap actually occurs, resulting in an inability to finely analyze local occlusion relationships; Second, a coherent technical path has not yet been formed to transform multi-layer sharpness variation information into pixel-level depth intervals, and further deduce the vertical relationships between cells and the global depth hierarchy structure; Third, traditional methods have poor stability across different samples and shooting conditions, often requiring frequent manual parameter adjustments, which can easily lead to inconsistencies or insensitivity to subtle occlusion relationships. Summary of the Invention

[0006] This invention proposes a cell occlusion relationship estimation method and system based on pixel-level focus evaluation curves, which solves the technical problems in existing cell microscopy imaging analysis techniques that cannot accurately estimate the depth range of a single pixel in cell overlap and boundary intersection areas, and lack a complete automated analysis path from multi-focal plane image sequences to pixel-level depth modeling, and then to the determination of local cell vertical relationships and global hierarchical sorting.

[0007] To address the aforementioned technical problems, this invention provides a method for estimating cell occlusion relationships based on pixel-level focus evaluation curves, comprising the following steps:

[0008] Step S1: Acquire multi-focal plane cell image sequences at different focal positions within the same field of view;

[0009] Step S2: Perform cell instance segmentation on the multi-focal plane cell image sequence to obtain a two-dimensional cell instance mask;

[0010] Step S3: For each pixel within the two-dimensional cell instance mask, calculate the sharpness value on each focal plane of the multi-focal plane cell image sequence, and construct a pixel-level focus response curve;

[0011] Step S4: Determine the focus depth range for each pixel based on the peak position of the pixel-level focus response curve and the preset relative threshold;

[0012] Step S5: In the cell overlap area, compare the focal depth ranges corresponding to pixels belonging to different cells to determine the vertical occlusion relationship between cells.

[0013] Preferably, step S1 further includes a preprocessing step: performing image registration, background subtraction, and brightness correction on the multi-focal plane cell image sequence.

[0014] Preferably, in step S2, the cell instance segmentation is achieved using one of the following methods: a threshold-based segmentation method, a morphology-based segmentation method, or a deep learning-based segmentation network.

[0015] Preferably, in step S3, the sharpness value is calculated using one of the following sharpness operators: Laplacian operator, Sobel operator, or Tenengrad operator.

[0016] Preferably, step S3 further includes: performing one-dimensional smoothing processing on the pixel-level focus response curve along the focal plane direction, wherein the one-dimensional smoothing processing adopts moving average filtering or Gaussian filtering.

[0017] Preferably, in step S4, the method for determining the focal depth range includes:

[0018] Step S4.1: Locate the peak position on the pixel-level focus response curve. and the corresponding maximum resolution value ;

[0019] Step S4.2: Set the relative threshold Calculate threshold sharpness = × ;

[0020] Step S4.3: Find a sharpness value greater than [value missing] on the pixel-level focus response curve. Within a continuous depth range, take the minimum depth. and maximum depth As the focal depth range [ , ].

[0021] Preferably, the relative threshold The value range is from 0.5 to 0.7.

[0022] Preferably, in step S5, the method for determining the vertical occlusion relationship between cells includes:

[0023] For two pixels belonging to the first cell and the second cell respectively at the same coordinate position within the cell overlap region, obtain the focal depth range of the corresponding pixel of the first cell. , ] and the focal depth range of the pixel corresponding to the second cell [ , ];

[0024] like < If so, it is determined that the first cell is located above the second cell;

[0025] like < If so, it is determined that the second cell is located above the first cell;

[0026] If two focal depth intervals overlap, the position is marked as uncertain.

[0027] Preferably, the method further includes:

[0028] For each pair of cells with overlapping areas, the vertical relationship of all pixels within the overlapping area is determined statistically, and the number of pixels above the first cell is calculated. The number of pixels above the second cell And the number of pixels is uncertain. ;

[0029] like > and / ( + )> If so, it is determined that the first cell as a whole is located above the second cell, wherein The threshold value is used.

[0030] The present invention also provides a cell occlusion relationship estimation system based on pixel-level focus evaluation curve, which is applicable to the above-mentioned method and includes an image acquisition module, a preprocessing module, a cell segmentation module, a focus evaluation module, and an occlusion determination module;

[0031] The image acquisition module is used to acquire multi-focal plane cell image sequences at different focus positions within the same field of view;

[0032] The preprocessing module is used to perform image registration, background subtraction, and brightness correction on the multi-focal plane cell image sequence;

[0033] The cell segmentation module is used to segment the multi-focal plane cell image sequence into cell instances to obtain a two-dimensional cell instance mask.

[0034] The focus evaluation module is used to construct a pixel-level focus response curve for each pixel within the two-dimensional cell instance mask, and determine the focus depth range of each pixel based on the pixel-level focus response curve.

[0035] The occlusion determination module is used to compare the focal depth ranges of pixels belonging to different cells in the cell overlap area and output the vertical occlusion relationship between cells.

[0036] The beneficial effects of the present invention include at least the following:

[0037] (1) The present invention introduces an expression method that combines pixel-level focus evaluation with depth range. Instead of giving a rough overall depth value for each cell, it calculates a specific and clear depth range for each pixel. This enables the precise differentiation of the vertical relationship between cells in complex areas such as cell boundary intersections and slight local overlaps, significantly improving the spatial resolution and reliability of the vertical relationship determination.

[0038] (2) This invention constructs a complete and practical analysis chain: starting from multi-focal plane continuous shooting images, the analysis area is first locked by segmenting stable cell instances, then the focal depth interval is established pixel by pixel, and finally local comparison is performed in the cell overlapping area and summarized into cell-level depth hierarchy, realizing the automatic conversion from raw microscopic data to global cell upper and lower layer relationship diagram, which greatly reduces the dependence on human experience and manual rules.

[0039] (3) The technical solution of the present invention is highly compatible with conventional microscopic imaging process. It can be deployed by simply adding a software analysis module on the basis of existing multi-focal plane imaging and cell segmentation. It does not rely on expensive three-dimensional reconstruction hardware and complex modeling, nor does it require manual annotation of regions of interest one by one, thus balancing engineering implementation cost and promotion value. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the occlusion relationship estimation method based on pixel-level focus evaluation curve in an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram of cell mask segmentation according to an embodiment of the present invention;

[0042] Figure 3 for Figure 2 A schematic diagram of the cell clarity-relative depth curves for two cells in the image;

[0043] Figure 4 for Figure 2 A schematic diagram showing the determination of the local occlusion relationship between two cells. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0045] like Figure 1 As shown, this embodiment of the invention provides a method for estimating cell occlusion relationships based on pixel-level focus evaluation curves, which includes the following steps.

[0046] Step S1: Acquire multi-focal plane cell image sequences at different focal positions within the same field of view.

[0047] In routine cell culture or tissue section imaging scenarios, multiple focused images are taken from the same field of view using an inverted microscope or confocal microscope. During imaging, parameters such as objective magnification, numerical aperture, and exposure time are kept constant, and the focal plane position is changed layer by layer along the optical axis in preset steps. One cell image is acquired at each focal plane, thus forming a sequence of multi-focal-plane cell images covering the sample thickness. Imaging conditions such as objective lens and exposure time are kept constant during the imaging process, and the focusing position corresponding to each image is recorded, ensuring that each image layer corresponds to a relative depth value.

[0048] After acquisition, the image sequence undergoes basic preprocessing. Preprocessing operations include: using registration algorithms to correct for any translation or slight rotation that may have occurred during imaging, ensuring strict alignment of each image layer on the plane; performing background subtraction and flat-field correction to reduce the effects of uneven illumination and background intensity drift; and, if necessary, smoothing or denoising of noisy data to improve the stability of subsequent focus evaluation. For researchers, this part of the process is consistent with the routine procedures currently used to obtain clear cell images and does not require changes to the existing experimental workflow.

[0049] Step S2: Perform cell instance segmentation on the multi-focal plane cell image sequence to obtain a two-dimensional cell instance mask.

[0050] Based on the preprocessed image sequence, a representative layer is selected as the reference image for two-dimensional cell segmentation. This reference image can be the result of maximum projection over all layers, a single intermediate layer, or a simple fusion image. Depending on the application requirements, threshold-based segmentation methods, morphology-based segmentation methods, or existing deep learning-based segmentation networks can be used to segment the reference image to obtain the foreground cell regions.

[0051] Subsequently, connected component analysis and watershed segmentation are used to separate contacting cells, generating an independent two-dimensional cell instance mask for each cell. These masks are then further processed with boundary smoothing, hole filling, and removal of small noise areas to obtain a stable and coherent set of cell instance masks, providing a reliable spatial range for subsequent pixel-level analysis. This invention treats these two-dimensional masks as the skeleton of the entire subsequent depth estimation: each mask corresponds to a cell instance, and the pixel position within each mask will obtain a depth range directly related to sharpness changes in subsequent steps, thus moving beyond coarse-grained information such as the approximate layer position of a single cell.

[0052] Step S3: For each pixel within the 2D cell instance mask, calculate the sharpness value on each focal plane of the multi-focal plane cell image sequence, and construct a pixel-level focus response curve.

[0053] Based on the cell segmentation results, this invention introduces a pixel-level focus evaluation step. Taking each pixel within the cell mask as a unit, the sharpness index of that pixel is calculated layer by layer across all focusing layers. Sharpness values ​​can be calculated using common sharpness operators such as the Laplacian operator, Sobel operator, or Tenengrad operator. Specifically, in each image layer, the aforementioned operators are applied to the neighborhood of the pixel, and the output response intensity is used as the focus score of that pixel on the corresponding focal plane. Thus, for a given pixel, a sequence of sharpness values ​​varying with the focal plane is obtained, which can be understood as the sharpness of the pixel on different focal planes.

[0054] To reduce the impact of random noise and single-layer outliers, a one-dimensional smoothing process is applied to the set of sharpness values ​​for the same pixel along the vertical axis. This one-dimensional smoothing can employ moving average filtering or Gaussian filtering, resulting in a smooth pixel-level focus response curve. This curve visually reflects the image quality of the pixel at different depths, with its peak typically corresponding to the sharpest image depth. Treating all these curves for all pixels as a single three-dimensional data volume constitutes the pixel-level focus evaluation volume.

[0055] Step S4: Determine the focus depth range of each pixel based on the peak position of the pixel-level focus response curve and the preset relative threshold.

[0056] After obtaining the focus response curve for each pixel, it is necessary to extract the depth range within which that pixel is likely to be in focus, i.e., the focus depth interval. The specific steps are as follows:

[0057] First, locate the peak position on the pixel-level focus response curve. and the corresponding maximum resolution value Then, set a relative threshold. Calculate threshold sharpness = × In this embodiment of the invention, the relative threshold The value typically ranges from 0.5 to 0.7 and can be adjusted based on the actual image quality and cell characteristics.

[0058] Finally, find the sharpness value greater than 1 on the pixel-level focus response curve. Take the minimum depth from the continuous depth range. and maximum depth As the focal depth range of this pixel [ , The lower bound of this interval can be understood as the shallowest point where the pixel just becomes clear, and the upper bound as the deepest point before it becomes blurry. The range between the two is the focal depth interval of the pixel. From an engineering perspective, only the minimum and maximum depth values ​​for each pixel can be recorded; when needed, the midpoint or peak depth of the interval can be used as a representative depth for quick visualization or to simplify subsequent calculations.

[0059] Based on the pixel-level focal depth range, a depth distribution map on the image plane is created for each cell. Specifically, within the area covered by the cell mask, each pixel is filled with the corresponding minimum and maximum sharpness depth (or representative depth value), and a vertical depth range is also included on the two-dimensional coordinate plane of the entire field of view. By superimposing the pixel-level depth information of all cells, a cell depth map is obtained, which reflects the vertical distribution of each cell at different positions in the entire field of view.

[0060] Step S5: In the cell overlap area, compare the focal depth ranges corresponding to pixels belonging to different cells to determine the vertical occlusion relationship between cells.

[0061] When different cells overlap on a plane, for the same coordinate position, the vertical relationship can be determined by directly comparing the depth range of each cell pixel. For example... Figures 2 to 4 As shown, for two pixels belonging to the first cell and the second cell respectively at the same coordinate position within the cell overlap area, after obtaining the focal depth range of the corresponding pixel in the first cell and the focal depth range of the corresponding pixel in the second cell, the following rules are applied for determination:

[0062] If the maximum value of the depth interval of the first cell The minimum value of the depth range less than the second cell If the difference exceeds a preset threshold, meaning the depth range of the first cell is generally closer to the lens than the second cell, then the first cell is determined to be above the second cell at that position. Conversely, if the maximum value of the depth range of the second cell is... The minimum value of the depth range less than the first cell If the two depth ranges overlap and it is difficult to clearly distinguish which is above the other, the pixel position is marked as uncertain to avoid misjudging noise or subtle differences as a stable occlusion relationship.

[0063] After local determination at the pixel level, the overall vertical relationship between cells is obtained by statistically analyzing the results of each pair of cells across all overlapping pixels. Specifically, for each pair of cells with overlapping areas on the plane, the number of pixels above the first cell is counted. The number of pixels above the second cell And the number of pixels is uncertain. .like Greater than ,and / ( + (greater than the threshold) If the first cell is significantly above the second cell, then the second cell is determined to be above the first cell. Conversely, if the second cell has a significantly dominant number of pixels above it, then the second cell is determined to be above the first cell. If both are close or weak, then the vertical relationship between the two cells is considered uncertain or insignificant. By repeating the above statistics for all cell pairs across the entire field of view, a deep hierarchical structure containing multiple vertical relationships can be constructed, thereby forming a vertical sorting or layering result of cells globally.

[0064] The present invention also provides a cell occlusion relationship estimation system based on pixel-level focus evaluation curves, including an image acquisition module, a preprocessing module, a cell segmentation module, a focus evaluation module, and an occlusion determination module.

[0065] The image acquisition module acquires multi-focal-plane cell image sequences from the same field of view at different focus positions. The preprocessing module performs preprocessing operations on the multi-focal-plane cell image sequences, including image registration, background subtraction, and brightness correction. The cell segmentation module segments the multi-focal-plane cell image sequences into two-dimensional cell instance masks. The focus evaluation module constructs a pixel-level focus response curve for each pixel within the two-dimensional cell instance mask and determines the focus depth range for each pixel based on the pixel-level focus response curve. The occlusion determination module compares the focus depth ranges corresponding to pixels belonging to different cells in overlapping cell regions and outputs the vertical occlusion relationship between cells.

[0066] The embodiments of the present invention are entirely based on conventional microscopic multi-focal plane images and existing cell segmentation results. They can be implemented simply by adding modules for focus evaluation calculation, depth interval calculation, and vertical relationship statistics at the software level, without requiring changes to existing imaging hardware or experimental procedures. Those skilled in the art can implement the method of the present invention on ordinary workstations or servers based on this specification and common image processing and numerical calculation libraries, and use the output cell depth maps and vertical relationship results for subsequent three-dimensional morphological analysis, cell contact pattern studies, or hierarchical distribution-based statistical calculations.

[0067] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described; only preferred embodiments of the present invention are illustrated. The descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. As long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0068] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this invention should be determined by the appended claims.

Claims

1. A method for estimating cell occlusion relationship based on pixel-level focus evaluation curve, characterized in that: It includes the following steps: Step S1: Collect a multi - focal - plane cell image sequence of the same field of view at different focus positions; Step S2: Perform cell instance segmentation on the multi - focal - plane cell image sequence to obtain a two - dimensional cell instance mask; Step S3: For each pixel in the two - dimensional cell instance mask, calculate the sharpness value on each focal plane of the multi - focal - plane cell image sequence, and construct a pixel - level focus response curve; Step S4: Determine the focus depth interval of each pixel according to the peak position of the pixel - level focus response curve and a preset relative threshold; The method for determining the focus depth interval includes: Step S4.1 : Locating the peak position on the pixel-level focus response curve and the corresponding maximum sharpness value ; Step S4.2: Setting a relative threshold , calculating a threshold sharpness = × ; Step S4.3: Find a sharpness value greater than [value missing] on the pixel-level focus response curve. Within a continuous depth range, take the minimum depth. and maximum depth As the focal depth range [ , ]; Step S5: In the cell overlapping area, compare the focus depth intervals corresponding to the pixels belonging to different cells to determine the upper - lower occlusion relationship between cells.

2. The method according to claim 1, characterized in that: In step S1, it further includes a pre - processing step: perform image registration, background subtraction, and brightness correction on the multi - focal - plane cell image sequence.

3. The method according to claim 1, characterized in that: In step S2, the cell instance segmentation is implemented by one of the following methods: a threshold - based segmentation method, a morphology - based segmentation method, a deep - learning - based segmentation network.

4. The method according to claim 1, characterized in that: In step S3, the sharpness value is calculated by one of the following sharpness operators: Laplacian operator, Sobel operator, Tenengrad operator.

5. The method according to claim 1, characterized in that: Step S3 further includes: perform one - dimensional smoothing on the pixel - level focus response curve along the focal - plane direction, and the one - dimensional smoothing uses moving average filtering or Gaussian filtering.

6. The method according to claim 1, wherein: The relative threshold The value range is from 0.5 to 0.

7.

7. The method according to claim 1, characterized in that: In step S5, the method for determining the upper - lower occlusion relationship between cells includes: For two pixels belonging to the first cell and the second cell respectively at the same coordinate position within the cell overlap region, obtain the focal depth range of the corresponding pixel of the first cell. , ] and the focal depth range of the pixel corresponding to the second cell [ , ]; like < If so, it is determined that the first cell is located above the second cell; If < , it is determined that the second cell is located above the first cell; If there is an overlap between two focus depth intervals, mark this position as uncertain.

8. The method according to claim 7, wherein The method further includes: For each pair of cells with an overlapping region, count the determination results of the up-and-down relationships of all pixels within the overlapping region, and calculate respectively the number of pixels where the first cell is above , the number of pixels where the second cell is above , and the number of pixels with uncertainty ; If > and / ( + ) > , it is determined that the whole of the first cell is located above the second cell, where is a threshold value.

9. A cell occlusion relationship estimation system based on a pixel-level focus evaluation curve, applicable to the method according to any one of claims 1 to 8, characterized in that: It includes an image acquisition module, a pre - processing module, a cell segmentation module, a focus evaluation module, and an occlusion determination module; The image acquisition module is used to collect a multi - focal - plane cell image sequence of the same field of view at different focus positions; The pre - processing module is used to perform image registration, background subtraction, and brightness correction on the multi - focal - plane cell image sequence; The cell segmentation module is used to perform cell instance segmentation on the multi - focal - plane cell image sequence to obtain a two - dimensional cell instance mask; The focus evaluation module is used to construct a pixel - level focus response curve for each pixel in the two - dimensional cell instance mask and determine the focus depth interval of each pixel according to the pixel - level focus response curve; The occlusion determination module is used to compare the focus depth intervals corresponding to the pixels belonging to different cells in the cell overlapping area and output the upper - lower occlusion relationship between cells.

Images