Cell density grouping method, device and equipment and storage medium

Abstract

The invention relates to the technical field of image processing, and particularly discloses a cell density grouping method, device and equipment and a storage medium, and the method comprises the following steps: carrying out gradient centrifugation and micro-fluidic chip combined homogenization treatment on a cell suspension to form a single-layer or three-dimensional hydrogel embedded sample; the method comprises the following steps: based on a single-layer / three-dimensional hydrogel embedding sample, collecting two-dimensional / three-dimensional dynamic image data of cells according to a time sequence mode by applying a high-resolution optical microscope and combining a fluorescence labeling technology; through combined homogenization treatment of gradient centrifugation and a micro-fluidic chip, and in combination with a high-resolution optical microscope and a fluorescence labeling technology, a cell dynamic image in a complex cell sample is efficiently and accurately captured, and a high-quality data basis is provided for subsequent analysis; by means of self-adaptive noise reduction processing and multi-scale morphological opening and closing operation, contour parameters, texture features and spatial topological relations of cells are accurately extracted, and the feature information lays a solid foundation for accurate distinguishing of cell subgroups.

Description

Technical Field

[0001] The present invention belongs to the field of image processing technology, and in particular relates to a cell density clustering method, device, equipment and storage medium. Background Art

[0002] In biomedical research and clinical diagnostics, accurate analysis of cell density and precise differentiation of cell subpopulations are crucial for understanding cellular function, disease progression, and evaluating drug efficacy. Traditional cell density clustering methods rely primarily on manual observation under a microscope and simple image processing techniques. These methods maintain a certain level of accuracy when processing cells with a simple structure and regular morphology.

[0003] As research deepens, the complexity of cell samples is increasing, especially when faced with high cell overlap, complex and changeable morphology, or in a dynamically changing state, the limitations of traditional methods are becoming increasingly prominent. In scenarios with high cell overlap, traditional image processing technology has difficulty in effectively separating adjacent cells, resulting in deviations in cell counts and density estimates. In addition, the complexity and dynamic variability of cell morphology, such as cell division, migration, or morphological changes, further increase the difficulty of accurately distinguishing cell subpopulations. These factors work together to seriously affect the accuracy of traditional methods in processing complex cell samples, thereby affecting the reliability and effectiveness of subsequent research. At the same time, with the development of high-throughput sequencing and single-cell analysis technologies, researchers are able to obtain larger and more complex cell image data sets. Traditional methods are not only inefficient when processing these large-scale data sets, but also difficult to ensure the consistency and repeatability of the analysis.

[0004] Therefore, it is necessary to propose a cell density clustering method, device, equipment and storage medium to solve the problem in the prior art that it is difficult to accurately distinguish cell subpopulations when processing complex cell samples.

[0005] The above information disclosed in this background technology is only for enhancing understanding of the background technology of the present invention and therefore it may contain information that does not constitute the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0006] The object of the present invention is to provide a cell density clustering method, device, equipment and storage medium to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A cell density clustering method comprising:

[0009] The cell suspension is homogenized by gradient centrifugation combined with a microfluidic chip to form a single layer or three-dimensional hydrogel-embedded sample;

[0010] Based on the monolayer / three-dimensional hydrogel-embedded sample, high-resolution optical microscopy combined with fluorescent labeling technology is used to collect two-dimensional / three-dimensional dynamic image data of cells in a time series mode;

[0011] Adaptively denoising the two-dimensional / three-dimensional dynamic image data, extracting cell contour parameters, texture features, and spatial topological relationships through multi-scale morphological opening and closing operations, and establishing a cell feature matrix containing position coordinates, morphological parameters, and adjacency relationships;

[0012] Based on the cell feature matrix, a cell spatial neighborhood network is constructed using the Delaunay triangulation technique based on deep graph embedding. Cell density clustering is performed in combination with the dynamic density-aware DBSCAN algorithm to generate initial cell density clustering results. The neighborhood radius is dynamically adjusted according to the local density gradient, and the initial clustering boundaries are probabilistically corrected using a graph convolutional network.

[0013] Based on the initial cell density clustering results, a dual-channel convolution-graph neural network model is constructed, in which the convolution branch processes the morphological feature map and the graph neural network branch processes the spatial topological relationship. The two features are fused through the cross-attention mechanism to output the cell subpopulation classification results with confidence scores.

[0014] Preferably, the dynamic density-aware DBSCAN algorithm introduces a density-sensitive distance measurement model based on deep learning, shrinking the search radius in high-density areas and expanding the neighborhood range in low-density areas;

[0015] Density-sensitive distance metric formula:

[0016]

[0017] Where, d adj (i, j) represents the density adaptive distance between cells i and j, x i and x j represents the spatial coordinates of cells i and j, ρ i and ρ j represents the k-nearest neighbor density centered on cells i and j.

[0018] Preferably, the dual-channel convolution-graph neural network model integrates self-supervised pre-training and online learning modules, uses contrastive learning and generative adversarial network technology to perform efficient feature learning on unlabeled data, and adapts to new data features through a continuous learning mechanism to optimize the generalization ability of the dual-channel convolution-graph neural network model.

[0019] Preferably, the method further provides a dynamic weighting mechanism for morphological features, and adjusts the contribution weights of the cell contour parameters and the texture features according to the cell overlap in the local area. The weight coefficient ω is calculated as follows:

[0020]

[0021] Where OI represents the region overlap index, and α represents the slope adjustment factor of the Sigmoid function.

[0022] Preferably, the method also generates an interactive three-dimensional spatial distribution heat map, simultaneously outputs a statistical report including cell subpopulation proportions, density gradient distribution, and morphological feature clustering, and displays the dynamic evolution of time series data;

[0023] The three-dimensional spatial distribution heat map integrates an optical tomography compensation algorithm to apply a density correction factor to deep cells to compensate for the effect of imaging depth on cell density distribution:

[0024] ρ corr =ρ obs ·e βz

[0025] Where, ρ corr represents the actual cell density after depth compensation, ρ obs represents the original observation density, z represents the vertical depth of the cell from the focal plane, and β represents the attenuation coefficient determined by the point spread function.

[0026] Preferably, the method further includes establishing a cross-modal and cross-sample comparison module, which aligns the cell density distribution characteristics between different sample types through deep transfer learning and optimal transmission theory to generate a standardized clustering map after batch effect correction.

[0027] A cell density clustering device, comprising:

[0028] The sample pretreatment module is used to homogenize the cell suspension through gradient centrifugation and microfluidic chip to form a single layer or three-dimensional hydrogel embedded sample;

[0029] An image acquisition module is configured to acquire two-dimensional / three-dimensional dynamic image data of cells in a time series mode based on the single-layer / three-dimensional hydrogel-embedded sample and configured with a high-resolution optical microscope combined with a fluorescent labeling component;

[0030] A feature extraction module is used to perform adaptive noise reduction processing on the two-dimensional / three-dimensional dynamic image data, and extract cell contour parameters, texture features and spatial topological relationships through multi-scale morphological opening and closing operations to generate a cell feature matrix containing position coordinates, morphological parameters and adjacency relationships;

[0031] Density clustering module, including:

[0032] A neighborhood network construction unit, configured to construct a cell space neighborhood network based on the cell feature matrix and a Delaunay triangulation technique based on depth map embedding;

[0033] Dynamic density clustering unit, used to perform cell density clustering using the dynamic density-aware DBSCAN algorithm to generate initial cell density clustering results, where the neighborhood radius is dynamically adjusted according to the local density gradient;

[0034] Boundary correction unit, used to perform probability correction on the initial cluster boundaries through graph convolutional networks;

[0035] A dual-channel fusion classification module is used to cluster the results based on the initial cell density, wherein the convolution branch processes the morphological feature map, and the graph neural network branch processes the spatial topological relationship. The two features are fused through the cross-attention mechanism to output the cell subpopulation classification results with confidence scores.

[0036] A cell density clustering device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement a cell density clustering method as described in any one of the above.

[0037] A cell density clustering storage medium having a computer program stored thereon, wherein the computer program, when executed by the processor, implements a cell density clustering method as described in any one of the above.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention uses a combined homogenization process of gradient centrifugation and microfluidic chips, combined with high-resolution optical microscopy and fluorescent labeling technology, to efficiently and accurately capture dynamic images of cells in complex cell samples, providing a high-quality data foundation for subsequent analysis; with the help of adaptive noise reduction processing and multi-scale morphological opening and closing operations, the contour parameters, texture features and spatial topological relationships of cells are accurately extracted. The accurate acquisition of these feature information lays a solid foundation for the accurate distinction of cell subpopulations; the Delaunay triangulation technology based on depth map embedding is used to construct a cell spatial neighborhood network, and the dynamic density-aware DBSCAN algorithm is combined to distinguish cell subpopulations of different densities in complex cell samples. In addition, based on the dual-channel convolution-graph neural network model, by fusing the morphological characteristics and spatial topological relationships of cells, the cell subpopulation classification results with confidence scores are output. This innovation not only optimizes the reliability of cell subpopulation classification, but also provides strong support for subsequent biomedical research and clinical diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a diagram of the cell density clustering method of the present invention;

[0041] Figure 2 This is a diagram of the cell density clustering device of the present invention. DETAILED DESCRIPTION

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0043] Example 1:

[0044] See also Figure 1 As shown, a cell density clustering method comprises:

[0045] The cell suspension is homogenized by gradient centrifugation combined with a microfluidic chip to form a single layer or three-dimensional hydrogel-embedded sample;

[0046] Based on single-layer / three-dimensional hydrogel-embedded samples, high-resolution optical microscopy combined with fluorescent labeling technology is used to collect two-dimensional / three-dimensional dynamic image data of cells in a time series mode;

[0047] Adaptive noise reduction is performed on 2D / 3D dynamic image data. Cell contour parameters, texture features, and spatial topological relationships are extracted through multi-scale morphological opening and closing operations to establish a cell feature matrix containing position coordinates, morphological parameters, and adjacency relationships.

[0048] Furthermore, by subjecting the cell suspension to gradient centrifugation and microfluidic chip homogenization to form a single-layer or three-dimensional hydrogel-embedded sample, and using high-resolution optical microscopy combined with fluorescent labeling technology to collect cell dynamic image data in a time series mode, the image data is then subjected to adaptive noise reduction and multi-scale morphological opening and closing operations to extract cell characteristic parameters, accurately obtaining the cell position coordinates, morphological parameters and adjacency relationships, providing a comprehensive and accurate cell feature matrix for subsequent cell density clustering and subpopulation classification, significantly improving the accuracy and reliability of cell analysis.

[0049] Based on the cell feature matrix, the Delaunay triangulation technique based on deep graph embedding is used to construct a cell spatial neighborhood network. The dynamic density-aware DBSCAN algorithm is then used to perform cell density clustering to generate the initial cell density clustering results. The neighborhood radius is dynamically adjusted according to the local density gradient, and the initial cluster boundaries are probabilistically corrected using a graph convolutional network.

[0050] The dynamic density-aware DBSCAN algorithm introduces a density-sensitive distance measurement model based on deep learning, which shrinks the search radius in high-density areas and expands the neighborhood range in low-density areas.

[0051] Furthermore, the Delaunay triangulation technology based on deep graph embedding was used to construct a cell space neighborhood network, and the dynamic density-aware DBSCAN algorithm that introduced a deep learning density-sensitive distance metric model was combined to perform cell density clustering. The neighborhood radius was dynamically adjusted according to the local density gradient to achieve a contracted search radius in high-density areas and an expanded neighborhood range in low-density areas, effectively improving the accuracy of cell density clustering. The initial clustering boundaries were probabilistically corrected through a graph convolutional network, further optimizing the accuracy and reliability of the clustering results.

[0052] Based on the initial cell density clustering results, a dual-channel convolution-graph neural network model is constructed. The convolution branch processes the morphological feature map, and the graph neural network branch processes the spatial topological relationship. The two features are fused through a cross-attention mechanism to output the cell subpopulation classification results with confidence scores.

[0053] The dual-channel convolutional-graph neural network model integrates self-supervised pre-training and online learning modules. It uses contrastive learning and generative adversarial network techniques to efficiently learn features from unlabeled data. It also adapts to new data features through a continuous learning mechanism, optimizing the generalization ability of the dual-channel convolutional-graph neural network model.

[0054] Furthermore, a dual-channel convolutional-graph neural network model was constructed, and self-supervised pre-training and online learning modules were integrated to efficiently fuse the morphological characteristics and spatial topological relationships of cells, and output cell subpopulation classification results with confidence scores, which significantly improved the accuracy and reliability of cell subpopulation classification; at the same time, contrastive learning and generative adversarial network technologies were used to efficiently learn features of unlabeled data, and a continuous learning mechanism was used to adapt to new data features, optimize the generalization ability of the model, and enable it to be more widely applied to the analysis of different cell samples.

[0055] This method also sets up a dynamic weighting mechanism for morphological features to adjust the contribution weights of cell contour parameters and texture features according to the cell overlap in the local area.

[0056] This method also generates interactive three-dimensional spatial distribution heat maps, and simultaneously outputs statistical reports including cell subpopulation proportions, density gradient distribution, and morphological feature clustering, and displays the dynamic evolution of time series data.

[0057] The three-dimensional spatial distribution heat map is integrated with an optical tomography compensation algorithm to apply a density correction factor to deep cells to compensate for the effect of imaging depth on cell density distribution.

[0058] The method also includes establishing a cross-modal and cross-sample comparison module, which aligns the cell density distribution characteristics between different sample types through deep transfer learning and optimal transfer theory to generate a standardized clustering map after batch effect correction.

[0059] Furthermore, the generated interactive three-dimensional spatial distribution heat map not only intuitively displays statistical information such as the proportion of cell subpopulations, density gradient distribution and morphological feature clustering, but also integrates an optical tomography compensation algorithm to correct the impact of imaging depth on cell density distribution, providing more accurate cell spatial distribution information; at the same time, the established cross-modal and cross-sample comparison module uses deep transfer learning and optimal transmission theory to effectively align the cell density distribution characteristics between different sample types, and generates a standardized clustering map after batch effect correction, which enhances the universality and accuracy of the cell density clustering method.

[0060] Example 2:

[0061] See also Figure 2 As shown, a cell density clustering device comprises:

[0062] The sample pretreatment module is used to homogenize the cell suspension through gradient centrifugation and microfluidic chip to form a single layer or three-dimensional hydrogel embedded sample;

[0063] Image acquisition module, used for collecting 2D / 3D dynamic image data of cells in a time series mode based on single-layer / 3D hydrogel-embedded samples and equipped with a high-resolution optical microscope combined with a fluorescent labeling component;

[0064] The feature extraction module is used to perform adaptive noise reduction on 2D / 3D dynamic image data and extract cell contour parameters, texture features, and spatial topological relationships through multi-scale morphological opening and closing operations to generate a cell feature matrix containing position coordinates, morphological parameters, and adjacency relationships;

[0065] Density clustering module, including:

[0066] The neighborhood network construction unit is used to construct the cell space neighborhood network based on the cell feature matrix and the Delaunay triangulation technology based on the depth map embedding;

[0067] Dynamic density clustering unit, used to perform cell density clustering using the dynamic density-aware DBSCAN algorithm to generate initial cell density clustering results, where the neighborhood radius is dynamically adjusted according to the local density gradient;

[0068] Boundary correction unit, used to perform probability correction on the initial cluster boundaries through graph convolutional networks;

[0069] A dual-channel fusion classification module is used to cluster the results based on the initial cell density. The convolution branch processes the morphological feature map, and the graph neural network branch processes the spatial topological relationship. The two features are fused through the cross-attention mechanism to output the cell subpopulation classification results with confidence scores.

[0070] Application example: Cell feature extraction and density clustering using cell density clustering technology

[0071] In biomedical research, researchers often need to extract cell features from complex cell samples and perform cell density clustering to gain a deeper understanding of cell function, behavior, and their role in disease progression. However, traditional cell analysis methods often struggle to efficiently and accurately extract cell features and perform effective density clustering when processing complex cell samples. Therefore, the introduction of advanced cell density clustering technology is particularly important.

[0072] 1. Application of technical solutions

[0073] This study employed an innovative cell density clustering method that combines gradient centrifugation with microfluidic chip homogenization, high-resolution optical microscopy, and fluorescent labeling to efficiently and accurately capture dynamic cell image data in complex cell samples. The following is a specific application example:

[0074] (1) Sample pretreatment:

[0075] The collected cell suspension was subjected to gradient centrifugation to remove impurities and concentrate the cells. The treated cell suspension was then combined with a microfluidic chip, where microfluidic technology was used to form a monolayer or three-dimensional hydrogel-embedded sample, providing a stable cell environment for subsequent image acquisition.

[0076] (2) Image acquisition:

[0077] Using high-resolution optical microscopy combined with fluorescence labeling, we acquire dynamic 2D / 3D image data of cells embedded in monolayer or 3D hydrogel-encapsulated samples in a time-series mode. The captured image data includes information such as cell morphology, position, and fluorescence signal at different time points.

[0078] (3) Image processing and feature extraction:

[0079] Adaptive noise reduction is performed on the collected image data to reduce noise interference and improve image quality. Multi-scale morphological opening and closing operations are used to accurately extract the cell contour parameters, texture features, and spatial topological relationships.

[0080] A cell feature matrix containing position coordinates, morphological parameters and adjacency relationships is established to provide basic data for subsequent density clustering.

[0081] (4) Cell density clustering:

[0082] Based on the cell feature matrix, a Delaunay triangulation technique based on depth map embedding was used to construct a cell spatial neighborhood network. The dynamic density-aware DBSCAN algorithm was then used to perform cell density clustering and generate the initial cell density clustering results.

[0083] During this process, the neighborhood radius is dynamically adjusted according to the local density gradient, and the initial clustering boundary is probabilistically corrected through the graph convolutional network to improve the accuracy of clustering.

[0084] (5) Cell subpopulation classification and result display:

[0085] Based on the initial cell density clustering results, a dual-channel convolutional-graph neural network model was constructed.

[0086] The convolution branch processes the morphological feature map, and the graph neural network branch processes the spatial topological relationship. The two features are fused through the cross-attention mechanism to output the cell subpopulation classification results with confidence scores.

[0087] Generate interactive 3D spatial distribution heat maps, simultaneously output statistical reports including cell subpopulation proportions, density gradient distribution, and morphological feature clustering, and display the dynamic evolution of time series data.

[0088] In addition, the three-dimensional spatial distribution heat map also integrates an optical tomography compensation algorithm to apply a density correction factor to deep cells to compensate for the effect of imaging depth on cell density distribution.

[0089] (6) Cross-modal and cross-sample comparison:

[0090] A cross-modality and cross-sample comparison module was established, using deep transfer learning and optimal transport theory to align cell density distribution characteristics across different sample types. This generated standardized cluster maps corrected for batch effects, enabling researchers to conduct comparative analyses across different samples.

[0091] 2. Application Effect

[0092] By applying this technical solution, biomedical experimenters can efficiently and accurately extract cell features from high-quality cell image data and perform effective density clustering. This not only improves the accuracy and reliability of cell analysis but also provides strong support for subsequent biomedical research and clinical diagnosis. Experimenters can gain a deeper understanding of cell function, behavior, and their role in disease progression, providing new ideas and methods for disease treatment and prevention.

[0093] Example 3:

[0094] An embodiment of the present invention further provides a cell density clustering device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement a cell density clustering method as described above.

[0095] The present invention also provides a cell density clustering storage medium having a computer program stored thereon. When the computer program is executed by a processor, it implements any of the cell density clustering methods described above and achieves the same technical effects. To avoid repetition, the details are not described here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0096] In the description of this specification, the reference terms "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0097] The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to common designs. In the absence of conflicts, the same embodiment and different embodiments of the present invention may be combined with each other.

[0098] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0099] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A cell density clustering method, characterized in that: include: The cell suspension is homogenized by gradient centrifugation combined with a microfluidic chip to form a single layer or three-dimensional hydrogel-embedded sample; Based on the monolayer / three-dimensional hydrogel-embedded sample, high-resolution optical microscopy combined with fluorescent labeling technology is used to collect two-dimensional / three-dimensional dynamic image data of cells in a time series mode; Adaptively denoising the two-dimensional / three-dimensional dynamic image data, extracting cell contour parameters, texture features, and spatial topological relationships through multi-scale morphological opening and closing operations, and establishing a cell feature matrix containing position coordinates, morphological parameters, and adjacency relationships; Based on the cell feature matrix, a cell spatial neighborhood network is constructed using the Delaunay triangulation technique based on deep graph embedding. Cell density clustering is performed in combination with the dynamic density-aware DBSCAN algorithm to generate initial cell density clustering results. The neighborhood radius is dynamically adjusted according to the local density gradient, and the initial clustering boundaries are probabilistically corrected using a graph convolutional network. Based on the initial cell density clustering results, a dual-channel convolution-graph neural network model is constructed, in which the convolution branch processes the morphological feature map and the graph neural network branch processes the spatial topological relationship. The two features are fused through the cross-attention mechanism to output the cell subpopulation classification results with confidence scores.

2. A cell density clustering method according to claim 1, characterized in that: The dynamic density-aware DBSCAN algorithm introduces a density-sensitive distance measurement model based on deep learning, which shrinks the search radius in high-density areas and expands the neighborhood range in low-density areas. Density-sensitive distance metric formula: Where, d adj (i, j) represents the density adaptive distance between cells i and j, x i and x j represents the spatial coordinates of cells i and j, ρ i and ρ j represents the k-nearest neighbor density centered on cells i and j.

3. A cell density clustering method according to claim 2, characterized in that: The dual-channel convolution-graph neural network model integrates self-supervised pre-training and online learning modules, uses contrastive learning and generative adversarial network technology to efficiently learn features of unlabeled data, and adapts to new data features through a continuous learning mechanism to optimize the generalization ability of the dual-channel convolution-graph neural network model.

4. A cell density clustering method according to claim 3, characterized in that: The method further sets a dynamic weighting mechanism for morphological features, and adjusts the contribution weights of the cell contour parameters and the texture features according to the cell overlap in the local area. The weight coefficient ω is calculated as follows: Where OI represents the region overlap index, and α represents the slope adjustment factor of the Sigmoid function.

5. A cell density clustering method according to claim 4, characterized in that: The method also generates interactive three-dimensional spatial distribution heat maps, and simultaneously outputs statistical reports including cell subpopulation proportions, density gradient distribution, and morphological feature clustering, and displays the dynamic evolution of time series data. The three-dimensional spatial distribution heat map integrates an optical tomography compensation algorithm to apply a density correction factor to deep cells to compensate for the effect of imaging depth on cell density distribution: r corr =ρ obs ·e βz Where, ρ corr represents the actual cell density after depth compensation, ρ obs represents the original observation density, z represents the vertical depth of the cell from the focal plane, and β represents the attenuation coefficient determined by the point spread function.

6. A cell density clustering method according to claim 5, characterized in that: The method also includes establishing a cross-modal and cross-sample comparison module, which aligns the cell density distribution characteristics between different sample types through deep transfer learning and optimal transmission theory to generate a standardized clustering map after batch effect correction.

7. A cell density clustering device, characterized in that: include: The sample pretreatment module is used to homogenize the cell suspension through gradient centrifugation and microfluidic chip to form a single layer or three-dimensional hydrogel embedded sample; An image acquisition module is configured to acquire two-dimensional / three-dimensional dynamic image data of cells in a time series mode based on the single-layer / three-dimensional hydrogel-embedded sample and configured with a high-resolution optical microscope combined with a fluorescent labeling component; A feature extraction module is used to perform adaptive noise reduction processing on the two-dimensional / three-dimensional dynamic image data, and extract cell contour parameters, texture features and spatial topological relationships through multi-scale morphological opening and closing operations to generate a cell feature matrix containing position coordinates, morphological parameters and adjacency relationships; Density clustering module, including: A neighborhood network construction unit, configured to construct a cell space neighborhood network based on the cell feature matrix and a Delaunay triangulation technique based on depth map embedding; Dynamic density clustering unit, used to perform cell density clustering using the dynamic density-aware DBSCAN algorithm to generate initial cell density clustering results, where the neighborhood radius is dynamically adjusted according to the local density gradient; Boundary correction unit, used to perform probability correction on the initial cluster boundaries through graph convolutional network; A dual-channel fusion classification module is used to cluster the results based on the initial cell density, wherein the convolution branch processes the morphological feature map, and the graph neural network branch processes the spatial topological relationship. The two features are fused through the cross-attention mechanism to output the cell subpopulation classification results with confidence scores.

8. A cell density clustering device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement a cell density clustering method according to any one of claims 1 to 6.

9. A cell density clustering storage medium, characterized in that: The storage medium stores a computer program, which, when executed by the processor, implements a cell density clustering method according to any one of claims 1 to 6.

Images