Method for distinguishing cell activity behaviors in culture medium under assistance of computer

Abstract

The invention provides a method for distinguishing cell activity behaviors in a culture medium under the assistance of a computer, and relates to the technical field of cell image processing, and the method comprises the following steps: obtaining a target microscopic image of a target cell at the current moment, extracting the static characteristics of the target cell in the target microscopic image, and if the recognition confidence is lower than a first threshold value, determining that the target cell is a cell activity behavior; if yes, calling a historical image sequence of an area where the target cell is located; at least N candidate images are obtained through screening in the historical image sequence, time sequence dynamic characteristics of the target cells are obtained based on the candidate images, and the time sequence dynamic characteristics comprise the motion trail, the motion speed, the morphological change related to the activity behavior and the periodic change condition presented by the target cells within a certain time period; and combining the time sequence dynamic characteristics with the static characteristics, and carrying out activity behavior matching to obtain activity behaviors of the target cells. According to the method, the robustness and the accuracy of cell behavior recognition in a complex culture environment are remarkably improved through static-dynamic feature fusion.

Description

Technical Field

[0001] The present application relates to the technical field of cell image processing, and in particular to a computer-assisted method for distinguishing cell activity behaviors in a culture medium. Background Art

[0002] In cell culture experiments, accurate identification of cell activity behaviors (such as division, apoptosis, and migration) is key to assessing cell status and microenvironmental responses.

[0003] In the past, the identification of cell activity behavior relied on single-frame static features. However, in the early stages of cell division or apoptosis, morphological changes may not be significant, which may cause misjudgment or missed detection. Therefore, it gradually evolved to combine static features with dynamic features to identify cell activity behavior.

[0004] However, in the process of combining dynamic features, dynamic features are mostly limited to trajectory tracking, and do not fully consider the spatial positioning stability of target cells (such as coordinate offset caused by culture medium flow) and the coordinated changes of the surrounding microenvironment. This in turn leads to insufficient fusion of dynamic and static features, and poor accuracy in distinguishing cell activity behaviors. Summary of the invention

[0005] In view of the above-mentioned defects or deficiencies in the prior art, the present application aims to provide a computer-assisted method for distinguishing cell activity behaviors in a culture medium, so as to improve the accuracy of distinguishing cell activity behaviors; The method comprises the following steps: Acquire a target microscopic image of a target cell at the current moment, and extract static features of the target cell in the target microscopic image, wherein the static features include the morphology, size, and texture of the cell; Determining the recognition confidence corresponding to the static feature of the target cell; if the recognition confidence is lower than a first threshold, retrieving a historical image sequence of the area where the target cell is located; Screening at least N candidate images in the historical image sequence, wherein the candidate images are historical images having similar spatial positioning features to the region where the target cell is located; the spatial positioning features include the positioning coordinates of the target cell and the distribution of other cells around the target cell; Based on the candidate image, the temporal dynamic characteristics of the target cell are obtained, wherein the temporal dynamic characteristics include the movement trajectory, movement speed, and morphological changes and periodic changes related to the activity behavior presented by the target cell within a certain period of time; The temporal dynamic features are combined with the static features to perform activity behavior matching to obtain the activity behavior of the target cell.

[0006] According to the technical solution provided by the present application, the temporal dynamic feature is combined with the static feature to perform activity behavior matching to obtain the activity behavior of the target cell, including the following steps: Combining the temporal dynamic feature with the static feature to obtain a fusion feature; Retrieving and traversing a behavior pattern library to obtain a reference feature that matches the fusion feature, and using the reference feature as a target reference feature, and using a reference behavior corresponding to the target reference feature as an activity behavior of the target cell; The behavior pattern library includes multiple cell types and a behavior pattern reference table for each cell type. The behavior pattern reference table includes multiple reference features and a reference behavior corresponding to each reference feature.

[0007] According to the technical solution provided by the present application, the step of screening at least N candidate images from the historical image sequence includes the following steps: Acquiring the location coordinates of the target cell from the target microscopic image; Extracting feature points of the target cell region in the target microscopic image, searching for matching points that match the feature points in each of the historical images, and taking the region covered by the matching points in the historical images as the cell corresponding region; Obtaining the cell coordinates in the area corresponding to the cell in each of the historical images, and obtaining the positioning coordinate distance corresponding to each of the historical images, wherein the positioning coordinate distance is the distance between the positioning coordinate and the cell coordinate; The historical image whose positioning coordinate distance is less than a first threshold is used as a first image, and a candidate image is obtained based on the first image.

[0008] According to the technical solution provided in this application, obtaining a candidate image based on the first image includes the following steps: In the target microscopic image, with the target cell as the center, the number and relative position relationship of other cells within a preset radius are counted to form a distribution of other cells; In each of the first images, taking the cell corresponding area as the center, counting the number and relative position relationship of other cells within the preset radius to form a first cell distribution of each of the first images; Obtaining a target feature vector corresponding to the other cell distributions and a first feature vector of each of the first cell distributions, and respectively calculating a cosine similarity between each of the first feature vectors and the target feature vector; The first image corresponding to the first feature vector whose cosine similarity is greater than a second threshold is taken as a candidate image.

[0009] According to the technical solution provided by the present application, after retrieving the historical image sequence of the area where the target cell is located, the following steps are also included: If the number of the candidate images screened in the historical image sequence is less than N, determining whether the total number of the historical images in the historical image sequence is a preset number, wherein the preset number is related to the time from the first introduction of the target cells into the culture dish to the present, and the acquisition frequency of the image acquisition device; If yes, the radius value of the preset radius range is expanded in batches, and the statistics of the other cell distributions and the first cell distribution are re-executed after each expansion to obtain the updated cosine similarity corresponding to each expansion; The historical images whose updated cosine similarity is greater than the second threshold are added as candidate images, and the above steps are repeated until the number of candidate images reaches N or the maximum number of expansion times is reached.

[0010] According to the technical solution provided by the present application, the historical image whose updated cosine similarity is greater than the second threshold is added as a candidate image, and then the following steps are further included: If the maximum number of expansion times is reached and the number of candidate images is less than N, performing missing dynamic feature extraction on all current candidate images to obtain missing dynamic features; Based on the missing dynamic features and the static features, the missing activity behavior of the target cell is obtained, and a confidence adjustment coefficient is added to the missing activity behavior, and the confidence adjustment coefficient is positively correlated with the number of candidate images.

[0011] According to the technical solution provided by the present application, the method for determining the maximum number of extensions comprises the following steps: Obtaining a current cell density value of the culture dish where the target cells are located; Determining a maximum allowable number of expansions according to the density-expansion mapping table and the current cell density value; The higher the current cell density value is, the smaller the corresponding maximum allowed expansion times are. The density-expansion mapping table is constructed by counting the cell distribution feature mismatch rates caused by adjacent expansion operations at different densities.

[0012] According to the technical solution provided in the present application, the step of extracting the static features of the target cells in the target microscopic image comprises the following steps: Divide an annular sampling area inside the contour of the target cell, wherein the annular sampling area includes at least three concentric rings; Calculating the histogram features of the local binary pattern for each of the annular bands; Performing a differential operation on the corresponding histogram features of two adjacent annular bands to generate a differential feature vector representing a texture gradient change; splicing the histogram features of each of the annular bands with the differential feature vector to obtain a two-dimensional texture feature; Based on the two-dimensional texture features, the texture of the target cell is obtained.

[0013] According to the technical solution provided by the present application, before extracting the static features of the target cells in the target microscopic image, the following steps are also included: Collecting multi-focal layer scanning data of the target cells, and constructing a three-dimensional voxel model of the target cells based on the multi-focal layer scanning data; Determining whether there is cell overlap among the target cells according to the three-dimensional voxel model; The step of obtaining the texture of the target cell based on the two-dimensional texture feature comprises the following steps: If so, separating the three-dimensional boundaries of the overlapping cells that overlap the target cell to generate a de-occluded surface topology of the target cell; The surface topological structure and the two-dimensional texture feature are cross-dimensionally fused to generate the texture of the target cell.

[0014] According to the technical solution provided in the present application, obtaining the temporal dynamic characteristics of the target cell based on the candidate image includes the following steps: If there is no cell overlap among the target cells, a two-dimensional motion speed is obtained based on the shooting time of each of the candidate images and the cell coordinates of the target cells, and the two-dimensional motion speed is used as the motion speed of the target cells; if there is cell overlap among the target cells, the spatial motion trajectory is extracted based on the three-dimensional voxel model, and the migration acceleration and directional deflection angle of the target cells in the three-dimensional space are calculated; and the two-dimensional motion speed is corrected according to the spatial motion trajectory, the migration acceleration and the directional deflection angle to obtain the motion speed of the target cells.

[0015] Compared with the prior art, the beneficial effects of the present application are as follows: when the static feature confidence is low (such as texture blurring caused by cell overlap), the present application ensures the spatiotemporal consistency of dynamic feature extraction by screening historical candidate images with similar spatial positioning features (positioning coordinates, surrounding cell distribution) as the current target cell, avoiding interference factors introduced by coordinate drift or drastic changes in the microenvironment; in addition, the temporal dynamic features not only include motion trajectories, but also cover motion speed, morphological changes and periodic changes related to active behaviors such as division and apoptosis. By fusing static features (current morphology) with dynamic features (morphological gradient trends), sub-dominant behavioral signals can be captured. For example, when the static confidence of the cell size is low, if it is detected that its motion speed decreases and the distribution of surrounding cells gradually becomes sparse (dynamic features), the apoptotic behavior of the cell can be accurately matched, greatly improving the accuracy of identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flowchart of the steps of the computer-assisted method for distinguishing cell activity behavior in culture medium provided in this application. DETAILED DESCRIPTION

[0017] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0018] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0019] Example 1 As mentioned in the background technology, in order to solve the problems in the prior art, this application proposes a computer-assisted method for distinguishing cell activity behavior in culture medium, such as Figure 1 As shown, the following steps are included: S1. Obtain a target microscopic image of a target cell at the current moment, and extract static features of the target cell in the target microscopic image, wherein the static features include cell morphology, size, and texture; Specifically, the images of the target cells in the culture dish are collected in real time by high-resolution microscopic imaging equipment (such as a confocal microscope). The image acquisition frequency is set to every 5 minutes / frame. The contours of the target cells are extracted using an image segmentation algorithm (such as a U-Net network), and the morphological parameters (such as area, perimeter, ellipticity) and size parameters (major axis, minor axis) are calculated; Furthermore, the extracting of static features of the target cells in the target microscopic image comprises the following steps: Divide an annular sampling area inside the contour of the target cell, wherein the annular sampling area includes at least three concentric rings; Calculating the histogram features of the local binary pattern for each of the annular bands; Performing a differential operation on the corresponding histogram features of two adjacent annular bands to generate a differential feature vector representing a texture gradient change; splicing the histogram features of each of the annular bands with the differential feature vector to obtain a two-dimensional texture feature; Based on the two-dimensional texture features, the texture of the target cell is obtained.

[0020] Specifically, three concentric rings (R1, R2, R3) are divided inside the contour of the target cell. The local binary pattern (LBP) histogram is calculated for each ring to obtain H1, H2, and H3. Adjacent rings are differentially operated (such as H1-H2, H2-H3) to generate differential vectors D1 and D2 that represent the texture gradient. H1-H3 and D1-D2 are concatenated into a two-dimensional texture feature matrix, and the texture feature vector of the target cell is obtained after dimensionality reduction through principal component analysis (PCA). At this point, the static feature is obtained.

[0021] S2, determining the recognition confidence corresponding to the static feature of the target cell; if the recognition confidence is lower than a first threshold, retrieving a historical image sequence of the area where the target cell is located; Specifically, static features are input into a pre-trained classifier (such as a support vector machine) to output the recognition confidence. The first threshold is set to 0.85. If the recognition confidence is less than 0.85, the historical image retrieval module is triggered to retrieve the historical image sequence (about 24 frames) of the target cell area in the past 2 hours and store it as a time stamp ordered queue.

[0022] S3, screening at least N candidate images in the historical image sequence, wherein the candidate images are historical images having similar spatial positioning features to the region where the target cell is located; the spatial positioning features include the positioning coordinates of the target cell and the distribution of other cells around the target cell; Furthermore, the step of screening at least N candidate images from the historical image sequence comprises the following steps: Acquiring the location coordinates of the target cell from the target microscopic image; Extracting feature points of the target cell region in the target microscopic image, searching for matching points that match the feature points in each of the historical images, and taking the region covered by the matching points in the historical images as the cell corresponding region; Obtaining the cell coordinates in the area corresponding to the cell in each of the historical images, and obtaining the positioning coordinate distance corresponding to each of the historical images, wherein the positioning coordinate distance is the distance between the positioning coordinate and the cell coordinate; The historical image whose positioning coordinate distance is less than a first threshold is used as a first image, and a candidate image is obtained based on the first image.

[0023] Further, obtaining a candidate image based on the first image comprises the following steps: In the target microscopic image, with the target cell as the center, the number and relative position relationship of other cells within a preset radius are counted to form a distribution of other cells; In each of the first images, taking the cell corresponding area as the center, counting the number and relative position relationship of other cells within the preset radius to form a first cell distribution of each of the first images; Obtaining a target feature vector corresponding to the other cell distributions and a first feature vector of each of the first cell distributions, and respectively calculating a cosine similarity between each of the first feature vectors and the target feature vector; The first image corresponding to the first feature vector whose cosine similarity is greater than a second threshold is taken as a candidate image.

[0024] Specifically, the coordinates (x0, y0) of the target cell in the current image are extracted, and the cell area with a Euclidean distance less than d=20 pixels from (x0, y0) is searched in the historical image to obtain 20 first images for preliminary screening. Then, in the target microscopic image, the coordinates of other cells within a radius of r=50μm are counted with the target cell as the center to generate other cell distribution vectors. For each first image, the first cell distribution vector of its corresponding area is calculated. The matching degree between the other cell distribution vectors corresponding to the target microscopic image and the first cell distribution vector corresponding to the first image is calculated by cosine similarity, and images with a similarity greater than the second threshold (such as 0.9) are retained as candidate images (N=10).

[0025] S4. Based on the candidate image, obtaining the temporal dynamic features of the target cell, wherein the temporal dynamic features include the movement trajectory, movement speed, and morphological changes and periodic changes associated with the activity behavior of the target cell within a certain period of time; Specifically, the temporal dynamic features also include the coordination of the distribution of target cells and surrounding cells, such as the appearance of gaps around cells. Time series analysis of 10 candidate images: Motion trajectory: The displacement of target cells between consecutive frames is tracked by the optical flow method, and the average motion speed v=Δs / Δt is calculated; where Δs represents the displacement difference of target cells between consecutive image frames (unit: micrometer), which is calculated by tracking the change of cell center of mass coordinates by the optical flow method. For example, between two adjacent frames (t and t+5 minutes), the center of mass coordinate changes from (x 1 ,y 1 ) moves to (x 2 ,y 2 ),but ; Δt represents the image acquisition time interval (unit: minutes), which is determined by the preset acquisition frequency. For example, when the acquisition frequency is every 5 minutes / frame, the adjacent frames Δt = 5 minutes.

[0026] Morphological changes: Detect the cell area change rate ΔA / A0, where A0 represents the initial area of ​​the target cell at the reference time (unit: square micrometer), usually the cell area of ​​the earliest frame in the historical image sequence (or the starting frame of the behavior analysis). ΔA represents the absolute value of the difference between the cell area at the current moment and A0; if ΔA increases by 40% within 10 minutes and is accompanied by an elongated ellipse, it is marked as a split feature. Synergy analysis: Statistical changes in the gaps between surrounding cells, if the gap area expansion rate is >5% / minute, it is determined to be a detachment phenomenon caused by apoptosis.

[0027] S5. Combining the temporal dynamic features with the static features to perform activity behavior matching to obtain the activity behavior of the target cell.

[0028] Furthermore, the temporal dynamic feature is combined with the static feature to perform activity behavior matching to obtain the activity behavior of the target cell, including the following steps: Combining the temporal dynamic feature with the static feature to obtain a fusion feature; Retrieving and traversing a behavior pattern library to obtain a reference feature that matches the fusion feature, and using the reference feature as a target reference feature, and using a reference behavior corresponding to the target reference feature as an activity behavior of the target cell; The behavior pattern library includes multiple cell types and a behavior pattern reference table for each cell type. The behavior pattern reference table includes multiple reference features and a reference behavior corresponding to each reference feature.

[0029] Specifically, the static features (morphological ellipticity 0.7, texture gradient peak 0.3) and the dynamic features ( , where v represents the two-dimensional movement speed of the cell, which is calculated by Δs / Δt. The features of the "mitotic precursor cells" (ellipticity 0.65-0.75, ΔA / A0>0.35) were matched in the behavior pattern library, and the behavior judgment results were output and the confidence level was marked as 0.92.

[0030] At this point, this embodiment completes the basic cell behavior recognition process. Because this implementation effectively combines the dynamic and static characteristics of the target cells, the accuracy of cell activity behavior recognition is greatly improved.

[0031] Example 2 Based on Example 1, this example can be applied to three-dimensional feature enhancement in cell overlapping scenarios. Considering that cells in a culture dish often overlap, this overlap will lead to difficulties in extracting static features of target cells and poor accuracy in analyzing temporal dynamic features. Therefore, this example is proposed to solve the above problems.

[0032] Before extracting the static features of the target cells in the target microscopic image, the following steps are also included: Collecting multi-focal layer scanning data of the target cells, and constructing a three-dimensional voxel model of the target cells based on the multi-focal layer scanning data; Determining whether there is cell overlap among the target cells according to the three-dimensional voxel model; Specifically, when the target cell overlaps with the neighboring cells in space, multi-focal layer scanning (layer spacing 1μm, 10 layers in total) is used to reconstruct the 3D voxel model through the deconvolution algorithm. The 3D boundary of the overlapping area is identified, and the surface topology of the target cell is separated using the active contour model.

[0033] The step of obtaining the texture of the target cell based on the two-dimensional texture feature comprises the following steps: If so, separating the three-dimensional boundaries of the overlapping cells that overlap the target cell to generate a de-occluded surface topology of the target cell; The surface topological structure and the two-dimensional texture feature are cross-dimensionally fused to generate the texture of the target cell.

[0034] Specifically, the two-dimensional texture features (LBP histogram) and the three-dimensional surface curvature features (based on the surface topology and calculated by the Poisson equation) are cascaded and input into the convolutional neural network to generate the fused texture descriptor. For example, the peak value of the two-dimensional texture gradient is 0.3, and the three-dimensional curvature change rate is 0.15. The fused features can distinguish the artifact interference between overlapping cells.

[0035] In a preferred embodiment, obtaining the temporal dynamic characteristics of the target cell based on the candidate image comprises the following steps: If there is no cell overlap among the target cells, a two-dimensional motion speed is obtained based on the shooting time of each of the candidate images and the cell coordinates of the target cells, and the two-dimensional motion speed is used as the motion speed of the target cells; if there is cell overlap among the target cells, the spatial motion trajectory is extracted based on the three-dimensional voxel model, and the migration acceleration and directional deflection angle of the target cells in the three-dimensional space are calculated; and the two-dimensional motion speed is corrected according to the spatial motion trajectory, the migration acceleration and the directional deflection angle to obtain the motion speed of the target cells.

[0036] Specifically, for overlapping cells, their motion trajectories in three-dimensional space are extracted. The migration acceleration in the z-axis direction (i.e., the depth direction of the culture dish) is calculated. If the migration acceleration in the z-axis direction is greater than 0.1 μm / s², the two-dimensional velocity is compensated. The compensation formula is: ,in, The three-dimensional motion speed corrected by the two-dimensional motion speed is the motion speed of the target cell. is the two-dimensional motion speed, is the migration acceleration in the z-axis direction, is the time interval, which is used to measure the time span from the beginning of recording the cell movement state to the current observation point during the observation of cell movement. At the same time, adjusting the movement direction vector according to the direction deflection angle can also improve the trajectory prediction accuracy.

[0037] Example 3 On the basis of Example 1, this embodiment proposes a solution to the situation where the number of candidate images may be insufficient (possible reasons for the insufficient number: if the target cell moves very fast and displaces greatly between different images, the number of images in the historical image whose positioning coordinates are less than the first threshold may be small, that is, the number of first images obtained is insufficient; or, the distribution of cells may be affected by various factors. If the environment in which the cells are located changes greatly, such as being stimulated by different chemical signals at different time points, the distribution of other cells around the target cells may vary greatly between different images; or, if the historical images have quality problems such as blurring and high noise, inaccurate feature point extraction or extraction failure may occur; or, at a certain moment, the image acquisition device fails and fails to successfully output the image at that moment, etc.), which may make subsequent operations unable to be performed.

[0038] After retrieving the historical image sequence of the area where the target cell is located, the following steps are also included: If the number of the candidate images screened in the historical image sequence is less than N, determining whether the total number of the historical images in the historical image sequence is a preset number, wherein the preset number is related to the time from the first introduction of the target cells into the culture dish to the present, and the acquisition frequency of the image acquisition device; If yes, the radius value of the preset radius range is expanded in batches, and the statistics of the other cell distributions and the first cell distribution are re-executed after each expansion to obtain the updated cosine similarity corresponding to each expansion; The historical images whose updated cosine similarity is greater than the second threshold are added as candidate images, and the above steps are repeated until the number of candidate images reaches N or the maximum number of expansion times is reached.

[0039] When the number of candidate images is less than N=10, the multi-level expansion mechanism is started: For example, the first expansion is performed: the radius r is expanded from 50μm to 70μm, the cell distribution similarity is recalculated, and 3 candidate images are added. The second expansion is performed: the radius is further expanded to 90μm, and 2 candidate images are added. Termination condition: the maximum number of expansions K=3 is reached (determined by checking the density-expansion mapping table based on the current cell density ρ=200 cells / mm²).

[0040] Furthermore, after the historical images whose updated cosine similarity is greater than the second threshold are added as candidate images, the following steps are also included: If the maximum number of expansion times is reached and the number of candidate images is less than N, performing missing dynamic feature extraction on all current candidate images to obtain missing dynamic features; Based on the missing dynamic features and the static features, the missing activity behavior of the target cell is obtained, and a confidence adjustment coefficient is added to the missing activity behavior, and the confidence adjustment coefficient is positively correlated with the number of candidate images.

[0041] For example, if only 8 candidate images are finally obtained, linear interpolation is performed on the missing 2 frames: based on the motion speeds v1~v8 of the existing frames, v9'=(v7+v8) / 2 is generated (estimated using the motion speeds of the two most recent frames (and) obtained), and v10'=v8+Δv (estimated using the motion speed and speed increment of the most recent frame obtained, and the speed increment is determined based on factors such as the change trend of the speed of the previous frames or the general law of cell movement), where v7 and v8 represent the cell movement speeds corresponding to the 7th and 8th candidate images in the historical image sequence (unit: micrometers per minute), which is the measured value; v9'=(v7+v8) / 2 represents the linear interpolation of the speed of the missing 9th frame, taking the arithmetic average of the speeds of the two most recent frames (v7, v8); v10'=v8+Δv: the extrapolated estimate of the speed of the missing 10th frame, where Δv represents the speed increment (unit: ), determined by one of the following two methods: extrapolated from the trend of the speed changes in the previous frames; calculated based on the historical speed sequence (such as v5~v8) to get the average acceleration. (a is the average acceleration, Δt is the frame interval); according to the general law of cell movement, the preset model is obtained: according to the default movement law of the cell type (for example, the typical acceleration range of migrating cells is 0.05-0.2 ). The interpolated features are combined with the measured features to obtain the missing activity behavior, which is then retrieved together with the static features and traversed through the behavior pattern library to obtain the matching result (missing activity behavior), and the confidence attenuation coefficient α=0.8 (α=actual number of frames / N) is added to the output result.

[0042] Furthermore, the method for determining the maximum number of extension times comprises the following steps: Obtaining a current cell density value of the culture dish where the target cells are located; Determining a maximum allowable number of expansions according to the density-expansion mapping table and the current cell density value; The higher the current cell density value is, the smaller the corresponding maximum allowed expansion times are. The density-expansion mapping table is constructed by counting the cell distribution feature mismatch rates caused by adjacent expansion operations at different densities.

[0043] Specifically, this embodiment proposes a solution for optimized screening under high-density culture. Through a large number of experiments or data statistics, the mismatch rate when performing expansion operations under different cell densities is obtained (mismatch rate β = number of mismatched cells / total number of cells), and then the maximum allowable number of expansions corresponding to different densities is determined to form a density-expansion mapping table.

[0044] For example, when the cell density ρ>500 cells / mm², the maximum number of expansions K=1 is limited according to the density-expansion mapping table. The distribution feature mismatch rate evaluation is introduced in the screening process: when β>0.2, the expansion is terminated to prevent the introduction of noise due to excessive radius expansion. At this time, the static feature weight enhancement strategy is preferentially adopted, and the static feature weight is adjusted from 0.5 to 0.7 during behavior matching.

[0045] Table 1

[0046] As shown in Table 1, the recognition accuracy of the traditional cell activity behavior identification method is compared with the method mentioned in this application. It can be seen that in the four different scenarios, the accuracy of the scheme mentioned in this application has been improved. Through static-dynamic feature fusion, three-dimensional modeling compensation and intelligent expansion mechanism, the robustness and accuracy of cell behavior recognition in complex culture environments are significantly improved.

[0047] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and its core ideas of this application. The above is only the preferred implementation method of this application. It should be pointed out that due to the limitations of textual expression and the objective existence of infinite specific structures, ordinary technicians in this technical field can make several improvements, modifications or changes without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes or combinations, or the direct application of the inventive concept and technical solution to other occasions without improvement, should be regarded as the scope of protection of this application.

Claims

1. A computer-assisted method for distinguishing cell activity behavior in a culture medium, characterized in that: The following steps are involved: Acquire a target microscopic image of a target cell at the current moment, and extract static features of the target cell in the target microscopic image, wherein the static features include the morphology, size, and texture of the cell; Determining the recognition confidence corresponding to the static feature of the target cell; if the recognition confidence is lower than a first threshold, retrieving a historical image sequence of the area where the target cell is located; Screening at least N candidate images in the historical image sequence, wherein the candidate images are historical images having similar spatial positioning features to the region where the target cell is located; The spatial positioning features include the positioning coordinates of the target cell and the distribution of other cells around the target cell; Based on the candidate image, the temporal dynamic characteristics of the target cell are obtained, wherein the temporal dynamic characteristics include the movement trajectory, movement speed, and morphological changes and periodic changes related to the activity behavior presented by the target cell within a certain period of time; The temporal dynamic features are combined with the static features to perform activity behavior matching to obtain the activity behavior of the target cell.

2. The computer-aided method for distinguishing cell activity behavior in culture medium according to claim 1, characterized in that: The step of combining the temporal dynamic features with the static features to perform activity behavior matching to obtain the activity behavior of the target cell comprises the following steps: Combining the temporal dynamic feature with the static feature to obtain a fusion feature; Retrieving and traversing a behavior pattern library to obtain a reference feature that matches the fusion feature, and using the reference feature as a target reference feature, and using a reference behavior corresponding to the target reference feature as an activity behavior of the target cell; The behavior pattern library includes multiple cell types and a behavior pattern reference table for each cell type. The behavior pattern reference table includes multiple reference features and a reference behavior corresponding to each reference feature.

3. The computer-aided method for distinguishing cell activity behavior in culture medium according to claim 1, characterized in that: The step of screening at least N candidate images from the historical image sequence comprises the following steps: Acquiring the location coordinates of the target cell from the target microscopic image; Extracting feature points of the target cell region in the target microscopic image, searching for matching points that match the feature points in each of the historical images, and taking the region covered by the matching points in the historical images as the cell corresponding region; Obtaining the cell coordinates in the area corresponding to the cell in each of the historical images, and obtaining the positioning coordinate distance corresponding to each of the historical images, wherein the positioning coordinate distance is the distance between the positioning coordinate and the cell coordinate; The historical image whose positioning coordinate distance is less than a first threshold is used as a first image, and a candidate image is obtained based on the first image.

4. The computer-aided method for distinguishing cell activity behavior in culture medium according to claim 3, characterized in that: The step of obtaining a candidate image based on the first image comprises the following steps: In the target microscopic image, with the target cell as the center, the number and relative position relationship of other cells within a preset radius are counted to form a distribution of other cells; In each of the first images, taking the cell corresponding area as the center, counting the number and relative position relationship of other cells within the preset radius to form a first cell distribution of each of the first images; Obtaining a target feature vector corresponding to the other cell distributions and a first feature vector of each of the first cell distributions, and respectively calculating a cosine similarity between each of the first feature vectors and the target feature vector; The first image corresponding to the first feature vector whose cosine similarity is greater than a second threshold is taken as a candidate image.

5. The computer-aided method for distinguishing cell activity behavior in culture medium according to claim 4, characterized in that: After retrieving the historical image sequence of the area where the target cell is located, the following steps are also included: If the number of the candidate images screened in the historical image sequence is less than N, determining whether the total number of the historical images in the historical image sequence is a preset number, wherein the preset number is related to the time from the first introduction of the target cells into the culture dish to the present, and the acquisition frequency of the image acquisition device; If yes, the radius value of the preset radius range is expanded in batches, and the statistics of the other cell distributions and the first cell distribution are re-executed after each expansion to obtain the updated cosine similarity corresponding to each expansion; The historical images whose updated cosine similarity is greater than the second threshold are added as candidate images, and the above steps are repeated until the number of candidate images reaches N or the maximum number of expansion times is reached.

6. The computer-aided method for distinguishing cell activity behavior in culture medium according to claim 5, characterized in that: After the historical image whose updated cosine similarity is greater than the second threshold is added as a candidate image, the following steps are also included: If the maximum number of expansion times is reached and the number of candidate images is less than N, performing missing dynamic feature extraction on all current candidate images to obtain missing dynamic features; Based on the missing dynamic features and the static features, the missing activity behavior of the target cell is obtained, and a confidence adjustment coefficient is added to the missing activity behavior, and the confidence adjustment coefficient is positively correlated with the number of candidate images.

7. The computer-aided method for distinguishing cell activity behavior in culture medium according to claim 5, characterized in that: The method for determining the maximum number of expansion times comprises the following steps: Obtaining a current cell density value of the culture dish where the target cells are located; Determining a maximum allowable number of expansions according to the density-expansion mapping table and the current cell density value; The higher the current cell density value is, the smaller the corresponding maximum allowed expansion times are. The density-expansion mapping table is constructed by counting the cell distribution feature mismatch rates caused by adjacent expansion operations at different densities.

8. The computer-aided method for distinguishing cell activity behavior in culture medium according to claim 1, characterized in that: The step of extracting the static features of the target cells in the target microscopic image comprises the following steps: Divide an annular sampling area inside the contour of the target cell, wherein the annular sampling area includes at least three concentric rings; Calculating the histogram features of the local binary pattern for each of the annular bands; Performing a differential operation on the corresponding histogram features of two adjacent annular bands to generate a differential feature vector representing a texture gradient change; splicing the histogram features of each of the annular bands with the differential feature vector to obtain a two-dimensional texture feature; Based on the two-dimensional texture features, the texture of the target cell is obtained.

9. The computer-aided method for distinguishing cell activity behavior in culture medium according to claim 8, characterized in that: Before extracting the static features of the target cells in the target microscopic image, the following steps are also included: Collecting multi-focal layer scanning data of the target cells, and constructing a three-dimensional voxel model of the target cells based on the multi-focal layer scanning data; Determining whether there is cell overlap among the target cells according to the three-dimensional voxel model; The step of obtaining the texture of the target cell based on the two-dimensional texture feature comprises the following steps: If so, separating the three-dimensional boundaries of the overlapping cells that overlap the target cell to generate a de-occluded surface topology of the target cell; The surface topological structure and the two-dimensional texture feature are cross-dimensionally fused to generate the texture of the target cell.

10. The computer-aided method for distinguishing cell activity behavior in culture medium according to claim 9, characterized in that: The step of obtaining the temporal dynamic characteristics of the target cell based on the candidate image comprises the following steps: If there is no cell overlap among the target cells, a two-dimensional motion speed is obtained based on the shooting time of each of the candidate images and the cell coordinates of the target cells, and the two-dimensional motion speed is used as the motion speed of the target cells; if there is cell overlap among the target cells, the spatial motion trajectory is extracted based on the three-dimensional voxel model, and the migration acceleration and directional deflection angle of the target cells in the three-dimensional space are calculated; and the two-dimensional motion speed is corrected according to the spatial motion trajectory, the migration acceleration and the directional deflection angle to obtain the motion speed of the target cells.

Images