Cell mechanical phenotype analysis method and equipment based on deep learning

Abstract

The invention is suitable for the technical field of cell analysis, and particularly relates to a cell mechanical phenotype analysis method and device based on deep learning, and the method comprises the following steps: obtaining a time sequence image, mechanical response data and culture environment parameters of a target cell; based on the mechanical response data and the culture environment parameters, performing region segmentation on the time sequence image, determining a mechanical sensitive region on the surface of the target cell and local deformation features of a subcellular structure, and realizing refined and targeted feature recognition; generating a mechanical phenotype descriptor of the target cell according to the mechanical sensitive area, the local deformation characteristics and the culture environment parameters to realize multi-modal data fusion; a mechanical phenotype analysis model obtained by training a sample mechanical phenotype descriptor of a sample cell in advance is used for processing the mechanical phenotype descriptor to obtain a heterogeneity prediction result of a target cell, the heterogeneity of the mechanical characteristics of the cell in a complex culture environment is automatically identified, and the dynamic nature and generalization ability of heterogeneity classification are remarkable.

Description

Technical Field

[0001] The present application belongs to the field of cell analysis technology, and in particular, relates to a cell mechanical phenotype analysis method and device based on deep learning. Background Art

[0002] Cell mechanical phenotype is the dynamic behavior of cells under external mechanical stimulation, which is closely related to cell function, pathological state and drug response. In recent years, with the deepening of biomechanical research, the analysis of cell mechanical properties has become an important research direction in the fields of disease diagnosis (such as cancer, cardiovascular disease), drug screening and regenerative medicine.

[0003] In the existing technology, heterogeneity analysis of cell mechanical phenotypes relies on single-point mechanical measurements (such as AFM) and artificial feature design, which requires pre-setting of static mechanical parameters and makes it difficult to dynamically track local deformations. It is also prone to misjudgment and missed detection in complex culture environments or at the subcellular scale. There is a lack of effective mechanical feature analysis tools, resulting in low heterogeneity classification accuracy and weak generalization. Summary of the invention

[0004] The embodiments of the present application provide a method and device for analyzing cell mechanical phenotypes based on deep learning, which can solve the problem of low heterogeneity classification accuracy and weak generalization due to the lack of effective mechanical feature analysis tools when analyzing heterogeneity of cell mechanical phenotypes.

[0005] In a first aspect, the present application provides a method for analyzing cell mechanical phenotypes based on deep learning, comprising: Acquire time-series images, mechanical response data, and culture environment parameters of target cells; Based on the mechanical response data and the culture environment parameters, the time series images are segmented to determine the mechanically sensitive areas on the surface of the target cells and the local deformation characteristics of the subcellular structures; generating a mechanical phenotype descriptor of the target cell according to the mechanically sensitive region, the local deformation characteristics and the culture environment parameters; The mechanical phenotype descriptor is processed using a mechanical phenotype analysis model to obtain a heterogeneity prediction result of the target cell; wherein the mechanical phenotype analysis model is a machine learning model pre-trained using a sample mechanical phenotype descriptor of a sample cell.

[0006] The above technical solutions in the embodiments of the present application have at least the following technical effects: The cell mechanical phenotype analysis method based on deep learning provided in the embodiment of the present application provides a fundamental data basis for subsequent analysis by acquiring the time series images, mechanical response data and culture environment parameters of the target cells. Based on the mechanical response data and culture environment parameters, the time series images are regionally segmented to determine the mechanically sensitive areas on the surface of the target cells and the local deformation characteristics of the subcellular structures, thereby achieving refined and targeted feature recognition. According to the mechanically sensitive areas, local deformation characteristics and culture environment parameters, the mechanical phenotype descriptors of the target cells are generated to achieve multimodal data fusion. The mechanical phenotype descriptors are processed using a mechanical phenotype analysis model pre-trained with a sample mechanical phenotype descriptor of a sample cell to obtain the heterogeneity prediction results of the target cells, automatically identify the heterogeneity of the mechanical properties of cells in a complex culture environment, and the dynamics and generalization capabilities of the significant heterogeneity classification.

[0007] In a second aspect, the present application provides a cell mechanical phenotype analysis system based on deep learning, comprising: An acquisition unit, used to acquire time-series images, mechanical response data and culture environment parameters of target cells; A segmentation unit, configured to perform regional segmentation on the time series image based on the mechanical response data and the culture environment parameters, and determine the mechanically sensitive area on the surface of the target cell and the local deformation characteristics of the subcellular structure; A generating unit, configured to generate a mechanical phenotype descriptor of the target cell according to the mechanically sensitive region, the local deformation characteristics and the culture environment parameters; A result unit is used to process the mechanical phenotype descriptor using a mechanical phenotype analysis model to obtain a heterogeneity prediction result of the target cell; wherein the mechanical phenotype analysis model is a machine learning model pre-trained using a sample mechanical phenotype descriptor of a sample cell.

[0008] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a method as described in any one of the above aspects when executing the computer program.

[0009] In a fourth aspect, an embodiment of the present application provides a computer program product. When the computer program product is run on an electronic device, the electronic device executes any one of the methods described in the above aspects.

[0010] It can be understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant descriptions in the above aspects and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0012] Figure 1 It is a schematic diagram of a process of a cell mechanical phenotype analysis method based on deep learning provided in one embodiment of the present application; Figure 2 This is a schematic diagram of the operation of a cell mechanical phenotype analysis method based on deep learning provided in an embodiment of the present application; Figure 3 It is a schematic diagram of the structure of a cell mechanical phenotype analysis system based on deep learning provided in one embodiment of the present application; Figure 4 It is a schematic diagram of the structure of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0013] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0014] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0015] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0016] As used in the specification of this application and the appended claims, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting" depending on the context. Similarly, the phrase "if it is determined" or "if the described condition or event is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce the described condition or event is detected" or "in response to detecting the described condition or event" depending on the context.

[0017] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0018] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0019] In the existing technology, heterogeneity analysis of cell mechanical phenotypes relies on single-point mechanical measurements (such as AFM) and artificial feature design, which requires pre-setting of static mechanical parameters and makes it difficult to dynamically track local deformations. It is also prone to misjudgment in complex culture environments or at the subcellular scale. The lack of effective mechanical feature analysis tools results in low heterogeneity classification accuracy and weak generalization.

[0020] To solve the above problems, the embodiments of the present application provide a method and device for analyzing cell mechanical phenotypes based on deep learning. In this method, by obtaining the time series images, mechanical response data and culture environment parameters of the target cells, a fundamental data basis is provided for subsequent analysis. Based on the mechanical response data and culture environment parameters, the time series images are regionally segmented to determine the mechanically sensitive areas on the surface of the target cells and the local deformation characteristics of the subcellular structures, so as to achieve refined and targeted feature recognition. According to the mechanically sensitive areas, local deformation characteristics and culture environment parameters, the mechanical phenotype descriptors of the target cells are generated to achieve multimodal data fusion. The mechanical phenotype descriptors are processed using the mechanical phenotype analysis model obtained by pre-training the sample mechanical phenotype descriptors of the sample cells to obtain the heterogeneity prediction results of the target cells, automatically identify the heterogeneity of the mechanical properties of cells in complex culture environments, and significantly classify the dynamics and generalization capabilities of heterogeneity.

[0021] The deep learning-based cell mechanical phenotype analysis method provided in the embodiment of the present application can be applied to electronic devices. In this case, the electronic device is the executor of the deep learning-based cell mechanical phenotype analysis method provided in the embodiment of the present application. The embodiment of the present application does not impose any restrictions on the specific type of electronic device.

[0022] For example, the electronic device may be an ultra-mobile personal computer (UMPC), a netbook, a desktop computer, a computer, a laptop computer, a communication device, a computing device, a satellite wireless device, etc.

[0023] In order to better understand the cell mechanical phenotype analysis method based on deep learning provided in the embodiment of the present application, the specific implementation process of the cell mechanical phenotype analysis method based on deep learning provided in the embodiment of the present application is exemplarily introduced below.

[0024] Figure 1 A schematic flow chart of a cell mechanical phenotype analysis method based on deep learning provided in an embodiment of the present application is shown. The cell mechanical phenotype analysis method based on deep learning includes: S100, acquiring time-series images, mechanical response data and culture environment parameters of target cells.

[0025] It can be understood that time-series images reflect the morphological changes of cells at different time points and can be obtained through high-resolution microscopes or live cell imaging systems; mechanical response data can be collected by devices such as microfluidic chips, atomic force microscopes or optical tweezers to record the stress-strain relationship of cells under mechanical stimulation; culture environment parameters include temperature, pH value, nutrient concentration, etc., which can be monitored in real time through sensors. Time-series images, mechanical response data and culture environment parameters together constitute the basic data set for cell mechanical behavior analysis, providing multi-dimensional data support for subsequent analysis, avoiding the limitations of a single data source, and improving the comprehensiveness and reliability of analysis results.

[0026] S200, based on the mechanical response data and the culture environment parameters, the time series images are segmented to determine the mechanically sensitive areas on the surface of the target cells and the local deformation characteristics of the subcellular structures.

[0027] It can be understood that the mechanical response data (such as stress distribution) and environmental parameters (such as culture medium hardness) can be integrated with image segmentation algorithms (such as U-Net and watershed algorithms) to divide the time series images into different functional areas and determine the mechanically sensitive areas on the cell surface. Mechanically sensitive areas refer to specific areas on the cell surface that are highly responsive to stress changes (such as cell membrane wrinkles and adhesion spots). The deformation characteristics of subcellular structures include morphological changes or movement trajectories of organelles such as mitochondria and endoplasmic reticulum.

[0028] In a possible implementation, S200, based on the mechanical response data and the culture environment parameters, performs regional segmentation on the time series image to determine the mechanically sensitive area on the surface of the target cell and the local deformation characteristics of the subcellular structure, including: S210, performing time-varying spatial registration processing on the time-series images to generate a time-varying spatial correlation relationship between the cell deformation field of the target cell and the motion trajectory of the subcellular structure.

[0029] It can be understood that the influence of cell movement or imaging drift can be eliminated through spatiotemporal alignment algorithms (such as dynamic time warping DTW) to establish the spatial correspondence between images at different time points. The cell deformation field describes the vector field of the overall shape change of the cell, and the subcellular structure motion trajectory records the curve of the position of the subcellular structure in three-dimensional space changing with time. The time-varying spatial correlation between the cell deformation field and the subcellular structure motion trajectory can be realized through coordinate mapping.

[0030] Optionally, S210, performing time-varying spatial registration processing on the time-series images to generate a time-varying spatial correlation relationship between the cell deformation field of the target cell and the motion trajectory of the subcellular structure includes: S211, performing displacement field calculation and feature matching on the time series images to determine the cell deformation field and subcellular structure motion trajectory of the target cell.

[0031] It can be understood that the displacement field calculation can use the optical flow method or block matching algorithm to track the displacement of pixels between adjacent images to determine the displacement field; feature matching uses algorithms such as SIFT and SURF to identify the feature points of the cytoskeleton or organelles and establish cross-frame correspondence. The deformation field is obtained by integrating the displacement field, and the motion trajectory of the subcellular structure is generated by fitting the feature point coordinate sequence.

[0032] Exemplarily, S211, performing displacement field calculation and feature matching on the time series images to determine the cell deformation field and subcellular structure motion trajectory of the target cell, includes: S2111, performing pixel-by-pixel displacement calculation on adjacent time-series images to generate an initial displacement field of the target cell.

[0033] It can be understood that the displacement vector of each pixel between adjacent frames can be calculated based on the Lucas-Kanade optical flow algorithm to obtain an initial displacement field reflecting the local movement of the cell. The initial displacement field quantifies the instantaneous deformation trend of the cell under mechanical stimulation.

[0034] S2112, identifying subcellular structure feature points from the time series image, and determining a motion trajectory coordinate sequence of the subcellular structure based on the subcellular structure feature points.

[0035] It can be understood that the characteristic points of the subcellular structure can be accurately located from the time-series images of the cell, and the coordinate sequence of its motion trajectory can be constructed based on these characteristic points, thereby realizing the tracking of the dynamic changes of the subcellular structure.

[0036] Exemplarily, when identifying the feature points of subcellular structures, the time series images can be preprocessed first, and Gaussian filtering can be used to remove noise in the image to avoid noise interfering with the identification of feature points. A grayscale-based feature point detection algorithm, such as the Harris corner detection algorithm, is used. The Harris corner detection algorithm determines corner points with significant features by calculating the grayscale changes of each pixel in the image in different directions. For subcellular structures, these corner points are often located at the edges, corners, etc. of the structure, and can effectively represent the characteristics of the subcellular structure. During the calculation process, the response value of each pixel point will be obtained, and a suitable threshold can be set in advance. Only pixels with a response value greater than the threshold will be confirmed as feature points. In this way, the feature points of subcellular structures can be preliminarily screened out in the time series images. The morphological features of subcellular structures can be combined. For example, mitochondria usually appear as elliptical or rod-shaped structures. Based on this feature, the detected feature points are further screened to remove those points that do not meet the mitochondrial morphological characteristics. The feature points that truly belong to mitochondria can be retained by calculating the shape factor, aspect ratio and other parameters of the area around the feature points and comparing them with the known range of mitochondrial morphological parameters. When determining the coordinate sequence of the motion trajectory of the subcellular structure, a method based on feature matching can be used. For two adjacent frames, the subcellular structure feature points are first detected in the first frame, and then matching feature points are found in the second frame. A matching algorithm based on feature descriptors is used, such as SIFT (Scale Invariant Feature Transform) descriptor or ORB (Oriented FAST and Rotated BRIEF) descriptor. Taking the ORB descriptor as an example, it has a certain invariance to the rotation and scale changes of the image, and has high computational efficiency. By calculating the ORB descriptor of the feature point in the first frame, searching for the most similar descriptor in the second frame, the matching feature point is found. In the matching process, a matching threshold can be set in advance, and only feature point pairs with a similarity higher than the threshold are considered to be valid matches. If the spatial position of a matching point in the second frame is too different from the corresponding point in the first frame, which exceeds the reasonable range of motion, the match is considered to be wrong and is eliminated. At the same time, the tracking window technology can be used. In the second frame image, a tracking window of a certain size is set with the position of the feature point in the first frame image as the center, and feature point matching is performed only within this window. This can reduce the amount of calculation, improve matching efficiency, and better cope with the overall movement and local deformation of cells.

[0037] When processing images of multiple subcellular structures, feature point matching conflicts may occur, that is, multiple feature points are matched to the same point in the second frame of the image. At this time, a conflict resolution strategy based on distance and similarity is used. The distance between each conflicting feature point and the matching point and the similarity of the descriptor are calculated, and the feature point with the closest distance and the highest similarity is selected as the correct matching point, thereby ensuring that the feature point of each subcellular structure can be accurately matched to the corresponding point in the next frame of the image, thereby generating an accurate motion trajectory coordinate sequence, identifying the subcellular structure feature points from the time series image, and determining its motion trajectory coordinate sequence, providing a reliable data basis for the subsequent analysis of the motion mode, deformation characteristics, and relationship with the overall mechanical response of the cell of the subcellular structure.

[0038] S2113, performing displacement correction based on the initial displacement field and the motion trajectory coordinate sequence of the subcellular structure to determine the cell deformation field and the subcellular structure motion trajectory of the target cell.

[0039] It can be understood that the acquired initial displacement field and subcellular structure motion trajectory coordinate sequence can be optimized to more accurately determine the cell deformation field and subcellular structure motion trajectory of the target cell. The acquired initial displacement field and subcellular structure motion trajectory coordinate sequence can be optimized using the least squares method. The motion trajectory coordinate sequence of the subcellular structure can be used as the actual observation value, and the predicted position of the subcellular structure calculated based on the initial displacement field can be used as the theoretical model prediction value. By minimizing the sum of squares of the errors between the two, the parameters in the initial displacement field are adjusted so that the adjusted initial displacement field can more accurately reflect the true movement of the subcellular structure.

[0040] Exemplarily, each coordinate point in the coordinate sequence of the motion trajectory of the subcellular structure can be used to calculate its predicted position in the current frame according to the initial displacement field. The error between the predicted position and the actual coordinate point is calculated, and the squares of all errors are accumulated. The parameters of the initial displacement field, such as the amplitude and direction of the displacement field, are adjusted continuously in an iterative manner so that the sum of square errors gradually decreases. When the sum of square errors reaches a preset threshold, it is considered that the displacement correction has achieved a satisfactory effect, and the optimized displacement field obtained at this time is the corrected displacement field. After obtaining the corrected displacement field, it is used to determine the cell deformation field of the target cell. The cell deformation field describes the deformation of the cell at different positions, which is closely related to the displacement field. By calculating the spatial derivative of the corrected displacement field, such as using the finite difference method, the rate of change of the displacement in different directions is calculated, thereby obtaining the cell deformation field. For example, the first-order derivatives of the displacement in the x-direction and the y-direction are calculated. These derivative information can reflect the deformation of the cell at various positions, such as stretching and compression, and then construct the cell deformation field. For the motion trajectory of the subcellular structure, optimization is performed on the basis of the corrected displacement field. According to the corrected displacement field, the coordinate sequence of the motion trajectory of the subcellular structure is adjusted. If the position of a subcellular structure is deviated due to errors in the initial motion trajectory coordinate sequence, the position will be corrected under the action of the corrected displacement field. The adjusted motion trajectory can be smoothed. For example, by using methods such as sliding average filtering, for each point in the motion trajectory coordinate sequence, the average value of several points before and after it is taken as the new coordinate value, so that the motion trajectory is smoother and more in line with the actual movement of the subcellular structure. Through the displacement correction operation of the initial displacement field and the subcellular structure motion trajectory coordinate sequence, the errors in the data can be effectively eliminated, and the cell deformation field of the target cell and the subcellular structure motion trajectory can be accurately determined.

[0041] S212, aligning the coordinates of the cell deformation field and the motion trajectory of the subcellular structure to establish a time-varying spatial mapping relationship between the cell deformation field and the subcellular structure.

[0042] It can be understood that after obtaining the cell deformation field and the motion trajectory of the subcellular structure, since they may be measured and calculated in different coordinate systems, or affected by factors such as the cell's own movement and changes in imaging angles, there is a lack of a unified spatial reference between the two, making it difficult to directly perform correlation analysis. Coordinate alignment operations can be performed to establish an accurate time-varying spatial mapping relationship between the two. The reference coordinate system can be constructed using the cell's center of mass or the position of a stable landmark subcellular structure (such as the nucleus) as a reference point. Transform the data of the cell deformation field and the subcellular structure motion trajectory. For the cell deformation field, its coordinates are converted to the selected reference coordinate system through geometric transformation operations such as translation, rotation, and scaling. The translation operation is to move the coordinate origin of the cell deformation field to the reference point position; the rotation operation is to rotate each point in the deformation field accordingly according to the overall rotation angle of the cell to eliminate the influence of cell rotation on the direction of the deformation field; the scaling operation is to unify the scale to ensure that the deformation field data at different time points or under different experimental conditions are comparable. For the subcellular structure motion trajectory, the above-mentioned geometric transformation is also performed. By calculating the relative position relationship between the motion trajectory of the subcellular structure and the reference point, its coordinates are adjusted to be consistent with the reference coordinate system. After completing the coordinate alignment, the time-varying spatial mapping relationship between the cell deformation field and the subcellular structure can be established. The time-varying spatial mapping relationship is established by matching each point on the motion trajectory of the subcellular structure with the point at the corresponding position in the cell deformation field at the same moment. For example, at a certain moment, the subcellular structure is located at a specific position on the motion trajectory. Through the coordinate information after coordinate alignment, the corresponding position in the cell deformation field is found, thereby determining the cell deformation experienced by the subcellular structure at this moment. In this way, the mapping from the motion trajectory of the subcellular structure to the cell deformation field is realized, and a time-varying spatial mapping relationship between the two is established.

[0043] S213, performing multi-scale strain analysis on the mapping relationship to generate a time-varying spatial correlation relationship between the cell deformation field of the target cell and the motion trajectory of the subcellular structure.

[0044] It can be understood that the core of multiscale strain analysis is to observe and quantify cell deformation and subcellular structure movement from different spatial and temporal scales. The mechanical behavior of cells may show different characteristics at different scales. At small scales, it may focus more on the local strain near the subcellular structure, while at large scales, it focuses on the strain trend of the whole cell.

[0045] For example, in terms of spatial scale, multi-resolution analysis methods such as wavelet transform can be used. Wavelet transform can decompose the data of cell deformation field and subcellular structure motion trajectory into components of different frequencies, and each frequency component corresponds to a different spatial scale. The low-frequency component reflects the overall change trend of the large scale, while the high-frequency component captures the detailed changes of the small scale. By analyzing the different frequency components, the strain of cells at different spatial scales can be studied separately. For example, for the cell deformation field, the low-frequency wavelet coefficients can show the macroscopic deformation characteristics of the whole cell, such as stretching and compression, while the high-frequency wavelet coefficients can reveal the subtle strain changes in local areas such as the cell edge and the surrounding of the subcellular structure. For the motion trajectory of the subcellular structure, wavelet analysis at different scales can help determine the differences in the motion patterns of the subcellular structure in the overall cell movement and local environmental changes.

[0046] Exemplarily, on the time scale, time series analysis methods such as the autoregressive moving average model (ARMA) can be used. The ARMA model can model the changes in the subcellular structure motion trajectory and the cell deformation field over time, and analyze their correlation and change trends at different time scales. Through the parameters of the model, rapid changes in a short period of time and slow evolution over a long period of time can be captured. Combining the analysis results of the spatial scale and the time scale, the time-varying spatial correlation relationship between the cell deformation field of the target cell and the subcellular structure motion trajectory is generated. The time-varying spatial correlation relationship not only includes the correspondence between cell deformation and subcellular structure motion at different spatial scales, but also reflects their mutual influence and change laws in the time dimension. For example, the motion response of the subcellular structure when the cell undergoes rapid local strain, as well as the long-term motion trend of the subcellular structure during the overall slow deformation of the cell, can be found.

[0047] S220, based on the time-varying spatial correlation relationship and the mechanical response data, the strain gradient tensor of the target cell surface is calculated to obtain the initial segmentation boundary of the mechanically sensitive area.

[0048] It is understandable that when cells are subjected to external mechanical stimulation or their own internal mechanical changes, different areas on their surface will produce different degrees of strain, and the strain gradient can reflect the severity of these strain changes. Mechanically sensitive areas are areas on the cell surface that are more sensitive to strain changes and respond strongly. The strain gradient tensor can be calculated by combining the time-varying spatial correlation (which integrates the morphological changes of cells in time and space dimensions and the movement information of subcellular structures) and mechanical response data (such as stress distribution, elastic modulus, etc.), and then the initial boundaries of the mechanically sensitive areas can be divided, which provides a key basis for in-depth research on the mechanical properties and functions of cells, and helps to understand how cells perceive and respond to mechanical signals, and the role of these mechanically sensitive areas in cell physiological and pathological processes.

[0049] Optionally, S220, based on the time-varying spatial correlation relationship and the mechanical response data, calculating the strain gradient tensor on the surface of the target cell to obtain the initial segmentation boundary of the mechanically sensitive area includes: S221, based on the time-varying spatial correlation relationship, obtain the strain tensor components of the target cell surface.

[0050] It can be understood that the time-varying spatial correlation records in detail the morphological changes of cells at different times and the movement trajectory of subcellular structures. This information reflects the deformation of cells in the space-time dimension. The strain tensor is a mathematical tool to describe the deformation state of an object. It can quantify the degree of tensile, compressive and shear deformation in different directions on the cell surface. Based on the time-varying spatial correlation, the strain tensor components can be calculated by analyzing the position changes of each point on the cell surface at different times using geometric and mechanical principles. For example, by tracking the displacement of specific marker points on the cell surface at different time points, combined with the relative position relationship between these points, and using tensor analysis methods, the normal strain components (measurement of tensile or compressive deformation) and shear strain components (measurement of shear deformation) in each direction are determined. The strain tensor components provide basic data for the subsequent calculation of the strain gradient tensor. They accurately describe the deformation state of the cell surface in different directions, which helps to further analyze the distribution characteristics of cell surface strain.

[0051] S222, performing spatial derivative operations on the strain tensor components on the target cell surface to generate a strain gradient tensor field.

[0052] It can be understood that the strain tensor components describe the strain state of each point on the cell surface, but to determine the severity and direction of the strain change, the strain gradient can be calculated. The spatial derivative operation can quantify the rate of change of the strain tensor components in space. Each strain tensor component (including the normal strain component and the shear strain component) is derived in each direction of space (such as the x, y direction, and the z direction if it is a three-dimensional cell). For example, the finite difference method is used to approximate the derivative of the strain tensor component in this direction by calculating the ratio of the difference between the strain tensor components at adjacent positions and the distance. After performing such operations on all strain tensor components and directions, these derivatives are combined to form the strain gradient tensor field. The strain gradient tensor field comprehensively reflects the intensity and direction information of the strain change on the cell surface, providing more detailed and accurate data support for determining the mechanically sensitive area. In the strain gradient tensor field, areas with larger values ​​indicate that the strain change is more severe. These areas are likely to be mechanically sensitive areas, and further analysis will be conducted based on this.

[0053] S223, determining a segmentation threshold of the mechanically sensitive area based on the strain gradient tensor field and the mechanical response data.

[0054] It can be understood that the strain gradient tensor field shows the distribution of strain changes on the cell surface, and the mechanically sensitive area can be divided by determining a suitable segmentation threshold. The mechanical response data contains various response information of cells to mechanical stimuli such as stress, such as the stress distribution of cells, changes in elastic modulus, etc. The segmentation threshold can be determined by the strain gradient tensor field and mechanical response data, so that the segmentation result is more consistent with the actual mechanical properties of the cell. For example, by analyzing a large amount of experimental data, the relationship between the strain gradient and the mechanical response of the cell can be found. For example, when the strain gradient reaches a certain value, the cell will produce a specific physiological response (such as changes in gene expression, activation of signal pathways, etc.). Statistical methods (such as histogram analysis and cluster analysis) can be used in combination with mechanical response data to determine a threshold that can effectively distinguish between mechanically sensitive areas and insensitive areas. In addition, machine learning methods can also be used to train the model using data samples of known mechanically sensitive areas and insensitive areas, so that the model can automatically learn and determine the optimal segmentation threshold. This segmentation threshold will serve as an important basis for the subsequent division of mechanically sensitive areas, which directly affects the accuracy of the final analysis results.

[0055] S224, dividing the surface area of ​​the target cell according to the segmentation threshold to obtain an initial segmentation boundary of the mechanically sensitive area.

[0056] It can be understood that after determining the segmentation threshold, the surface area of ​​the target cell can be divided according to this threshold, so as to obtain the initial segmentation boundary of the mechanically sensitive area. The strain gradient value of each point in the strain gradient tensor field is compared with the segmentation threshold. If the strain gradient value of a point is greater than the segmentation threshold, the area where the point is located is marked as a possible mechanically sensitive area; otherwise, it is marked as an insensitive area. Through such a comparison and marking process, the cell surface is divided into different areas. Then, the boundaries are extracted from these marked areas using image processing and boundary extraction algorithms (such as contour detection algorithms). For example, the Canny edge detection algorithm is used, which can detect the edge of the object according to the change of grayscale value in the image (here, the change of strain gradient value), so as to obtain the initial segmentation boundary of the mechanically sensitive area. The initial segmentation boundary preliminarily defines the scope of the mechanically sensitive area, and provides a basic framework for further studying the characteristics and functions of the mechanically sensitive area and its relationship with other parts of the cell.

[0057] S230, optimizing the initial segmentation boundary according to the strain gradient tensor and the culture environment parameters, and determining the mechanically sensitive area on the surface of the target cell.

[0058] It is understandable that after obtaining the initial segmentation boundary of the mechanically sensitive area, since the culture environment in which the cells are located will affect the mechanical properties of the cells, and the initial segmentation boundary may be inaccurate and subject to noise interference, it is necessary to optimize it in combination with the strain gradient tensor and the culture environment parameters to more accurately determine the mechanically sensitive area. The strain gradient tensor reflects the severity of the strain change on the cell surface, while the culture environment parameters (such as temperature, pH value, culture medium composition, etc.) will change the physiological state and mechanical response of the cells. The two can be comprehensively considered to improve the accuracy and reliability of the determination of the mechanically sensitive area, which is of great significance for in-depth research on the mechanical behavior of cells in different environments.

[0059] Optionally, S230, optimizing the initial segmentation boundary according to the strain gradient tensor and the culture environment parameters to determine the mechanically sensitive area on the surface of the target cell, includes: S231, dynamically adjusting the segmentation threshold of the initial segmentation boundary based on the culture environment parameters to generate an optimized threshold of the mechanically sensitive area.

[0060] It can be understood that changes in culture environment parameters can affect the mechanical properties of cells, thereby changing the distribution and characteristics of mechanically sensitive areas. For example, changes in temperature may cause changes in the fluidity of the cell membrane, affecting the cell's response to mechanical stimulation; differences in culture medium components may affect the metabolic activity of cells and the stability of the cytoskeleton, thereby changing the mechanical properties of cells. Based on this, it is necessary to dynamically adjust the segmentation threshold set when the initial segmentation boundary is determined according to the culture environment parameters. Statistical analysis, machine learning and other methods can be used to establish a regression model (such as multivariate linear regression) between environmental parameters (such as pH value, temperature) and strain gradient thresholds to determine the influence of different culture environment parameters (such as temperature, pH value, concentration of specific nutrients, etc.) on cell strain gradients. For example, it is found that when the temperature rises, the elastic modulus of the cell decreases, and the response to stress is more sensitive, and the strain gradient threshold of the mechanically sensitive area may need to be adjusted accordingly. According to the current culture environment parameters, the initial segmentation threshold is adjusted according to the above relationship model. If the temperature in the current culture environment is high, according to the model prediction, the cell's response to mechanical stimulation is enhanced, so the segmentation threshold is appropriately lowered to include more parts with relatively low strain gradients but may still be mechanically sensitive areas in the current environment; conversely, if the temperature is low, the cell's response to mechanical stimulation is weakened, so the segmentation threshold is appropriately increased to exclude some parts that may have abnormal strain gradients due to environmental factors but are not truly mechanically sensitive areas. Through dynamic adjustment, an optimized threshold for mechanically sensitive areas that is more in line with the actual situation of cells in the current culture environment is generated.

[0061] S232, performing morphological closing operation on the initial segmentation boundary according to the optimized threshold value to generate an optimized boundary of the mechanical sensitive area.

[0062] It can be understood that the morphological closing operation can be used to process the initial segmentation boundary to eliminate noise and fill holes, so as to make the boundary of the mechanical sensitive area more accurate and continuous. The morphological closing operation consists of two basic morphological operations: dilation and erosion.

[0063] Exemplarily, the dilation operation is to expand the boundary of the object in the image outward. For the mechanically sensitive area defined by the initial segmentation boundary, the dilation operation can fill the boundary holes caused by noise or inaccurate segmentation and connect some discrete small areas. For example, if there are some small holes in the initial segmentation boundary, it may be caused by the measurement error of the local strain gradient or noise interference. The dilation operation can fill these holes to make the shape of the mechanically sensitive area more complete. The erosion operation shrinks the boundary of the object inward. Performing an erosion operation after dilation can remove some false edges and noise points introduced by the dilation, making the boundary smoother and more accurate. The quality of the initial segmentation boundary can be effectively improved by combining dilation and erosion (i.e., closing operation). Dilation and erosion operations can be performed through suitable structural elements (such as circles, squares, etc.). By performing morphological closing operations on the initial segmentation boundary, a more accurate and more representative optimized boundary of the actual mechanically sensitive area is generated.

[0064] S233, performing boundary fusion on the optimized boundary according to the strain gradient tensor to determine the mechanically sensitive area on the surface of the target cell.

[0065] It can be understood that the strain gradient change trend on both sides of the optimization boundary can be identified in the strain gradient tensor. If there is a significant difference in the strain gradient on both sides of the boundary, and this difference conforms to the characteristics of the mechanically sensitive area, it means that the current boundary division may be reasonable; but if the strain gradient change is not obvious or there is an abnormality, the boundary may need to be adjusted. Boundary fusion can be performed based on the information of the strain gradient tensor. For adjacent parts that may belong to the mechanically sensitive area, if the strain gradient change between them is continuous and conforms to the characteristics of the mechanically sensitive area, the boundaries of these parts are fused to form a larger and more continuous mechanically sensitive area. For example, there are some small segments of boundaries on the optimization boundary, and the strain gradient changes in the areas defined by them are similar and significantly different from the surrounding areas. These small segments of boundaries can be merged through boundary fusion to make the boundaries of the mechanically sensitive area more coherent.

[0066] S240, extracting the motion trajectory of the subcellular structure from the time-varying spatial correlation relationship, calculating the curvature change parameter of the subcellular structure, and determining the local deformation characteristics of the subcellular structure.

[0067] It is understandable that the movement and deformation of subcellular structures play a key role in the physiological activities of cells. Studying their movement trajectories and local deformation characteristics helps to gain a deeper understanding of cell functions and physiological processes. The time-varying spatial correlation integrates the information of cells in time and space dimensions, providing a rich data basis for extracting the movement trajectories of subcellular structures. By calculating the curvature change parameters of subcellular structures, the changes in the curvature of their movement trajectories can be quantified, thereby determining the local deformation characteristics. The local deformation characteristics can reflect the morphological changes of subcellular structures during movement, which is of great significance for revealing the mechanical mechanisms and physiological functions within cells.

[0068] Optionally, S240, extracting the motion trajectory of the subcellular structure from the time-varying spatial correlation relationship, calculating the curvature change parameter of the subcellular structure, and determining the local deformation characteristics of the subcellular structure, includes: S241, extracting the motion trajectory of the subcellular structure from the time-varying spatial correlation relationship, and determining the motion trajectory coordinate sequence of the subcellular structure of the target cell.

[0069] It can be understood that the time-varying spatial correlation relationship records in detail the spatial position information of the subcellular structure at different times and its association with the overall deformation of the cell. The parts related to the movement of the subcellular structure can be screened out from the time-varying spatial correlation relationship, and the specific position of the subcellular structure at each time point can be obtained, and then the coordinate sequence of its motion trajectory can be determined. Since the time-varying spatial correlation relationship may contain a large amount of redundant information, specific algorithms and rules can be used to extract useful data. For example, according to the characteristic mark of the subcellular structure or a pre-set recognition method, the position of the subcellular structure can be accurately found in the spatial data at different time points. For subcellular structures such as mitochondria, their specific fluorescent markers can be used to locate the position of mitochondria in the image data corresponding to the time-varying spatial correlation relationship through image processing technology. Arranging the position information of each time point in chronological order forms the motion trajectory coordinate sequence of the subcellular structure. This coordinate sequence accurately records the position changes of the subcellular structure at different times, providing basic data for the subsequent analysis of its motion characteristics and deformation characteristics.

[0070] S242, performing spline interpolation on the motion trajectory coordinate sequence to generate a motion trajectory curve of the subcellular structure, and calculating a trajectory curvature parameter of the subcellular structure according to the motion trajectory curve.

[0071] It can be understood that the obtained coordinate sequence of the motion trajectory of the subcellular structure is usually discrete data points, and these discrete points are difficult to intuitively reflect the continuity and smoothness of the subcellular structure movement. Spline interpolation is a commonly used mathematical method, through which a smooth curve can be constructed between these discrete coordinate points to more accurately describe the motion trajectory of the subcellular structure. The principle of spline interpolation is to use piecewise polynomial functions to fit discrete data points so that the curve maintains a certain smoothness while passing through each data point. Spline interpolation methods include cubic spline interpolation, etc. Cubic spline interpolation ensures that the curve has continuous first-order and second-order derivatives in the entire interval by constructing a cubic polynomial function between each data point, thereby ensuring the smoothness of the curve. In practical applications, a suitable spline interpolation method is selected according to the characteristics of the motion trajectory coordinate sequence to generate the motion trajectory curve of the subcellular structure. After obtaining the motion trajectory curve, the trajectory curvature parameter can be calculated to quantify the curvature of the curve. Curvature is an important parameter to describe the curvature of the curve. For the motion trajectory curve of the subcellular structure, the curvature parameter can reflect the curvature change during the motion process. There are many ways to calculate curvature. For example, for parameterized curves, the first and second derivatives of the curve can be used to calculate the curvature. By calculating the trajectory curvature parameters, we can intuitively understand the degree of curvature of the motion trajectory of subcellular structures at different positions and times, and provide a quantitative basis for analyzing its local deformation characteristics.

[0072] S243, performing a time derivative operation according to the trajectory curvature parameter of the subcellular structure to determine a local deformation feature of the subcellular structure. It can be understood that the local deformation of the subcellular structure is not only related to the curvature of the motion trajectory, but also closely related to the change of curvature over time. By performing time derivative operations on the trajectory curvature parameters, the rate of change of curvature over time can be obtained, so as to more accurately determine the local deformation characteristics of the subcellular structure. The time derivative operation reflects the changing trend of the trajectory curvature in the time dimension. When the subcellular structure undergoes local deformation, the curvature of its motion trajectory will change over time, and this change can be more clearly reflected by the time derivative operation. For example, when mitochondria are in the process of fission or fusion, the curvature of their motion trajectory will change rapidly. After performing time derivative operations on the curvature parameters, the rate and direction of this change can be captured. According to the results of the time derivative operation, the local deformation characteristics of the subcellular structure can be determined. If the time derivative of the curvature is large, it means that the curvature of the subcellular structure has changed significantly in a short period of time, which may mean that the subcellular structure is undergoing a more drastic local deformation, such as the constriction or elongation of mitochondria; on the contrary, if the time derivative is small, it means that the curvature of the subcellular structure changes slowly and the local deformation is relatively small. By taking the time derivative of the trajectory curvature parameter as an important indicator, the local deformation characteristics of the subcellular structure can be effectively determined, providing key information for in-depth research on the dynamic changes of subcellular structures and the physiological processes of cells.

[0073] S300, generating a mechanical phenotype descriptor of the target cells based on the mechanically sensitive regions, local deformation characteristics, and culture environment parameters.

[0074] It can be understood that the mechanical phenotype of a cell is a manifestation of its comprehensive characteristics under a mechanical environment, reflecting the cell's response to mechanical stimulation and its own mechanical properties. The mechanically sensitive region reflects the part of the cell surface that is sensitive to mechanical stimulation, the local deformation characteristics show the changes in the subcellular structure under mechanical action, and the culture environment parameters affect the physiological state and mechanical behavior of the cell. The mechanical phenotype descriptor is generated by integrating the information of the mechanically sensitive region, local deformation characteristics and culture environment parameters, which can comprehensively and quantitatively characterize the mechanical phenotype of the cell, provide a unified quantitative indicator for subsequent research on the mechanical properties, functions and changes of cells in different environments, and help to deeply understand the physiological and pathological processes of cells, as well as the mechanical heterogeneity between cells.

[0075] In a possible implementation, S300 generates a mechanical phenotype descriptor of the target cell according to the mechanically sensitive region, the local deformation characteristics, and the culture environment parameters, including: S310, performing feature association according to the mechanically sensitive area, the local deformation characteristics and the culture environment parameters to generate a heterogeneous association matrix of the target cells.

[0076] It can be understood that there are complex relationships between the mechanically sensitive areas of cells, local deformation characteristics, and culture environment parameters, which affect the mechanical phenotype and heterogeneity of cells. The purpose of feature association is to explore the intrinsic connections between these factors. By analyzing the correlation, causal relationship, etc. between different factors, a matrix is ​​constructed to represent the degree of association between them, namely the heterogeneity association matrix.

[0077] For example, the characteristics of mechanically sensitive areas (such as area, position, strain gradient, etc.), local deformation characteristics (such as the curvature change rate of subcellular structures, deformation amplitude, etc.), and culture environment parameters (temperature, pH value, nutrient concentration, etc.) can be used as the element dimensions of the matrix. Calculate the degree of correlation between the elements, such as using the Pearson correlation coefficient to measure the linear correlation between two features. If the change trends of the two features are similar, the correlation coefficient is high, and the corresponding element value in the matrix is ​​also large; conversely, if there is no obvious correlation between the two features, the element value is small. In this way, a heterogeneous correlation matrix that can fully reflect the correlation between the factors is generated. The matrix provides a basic data structure for the subsequent in-depth analysis of cell heterogeneity.

[0078] S320, performing feature dimensionality reduction on the heterogeneous association matrix, and performing time-frequency joint feature encoding on the heterogeneous association matrix after dimensionality reduction to generate a mechanical phenotype descriptor of the target cell.

[0079] It can be understood that the heterogeneous association matrix contains the association information between multiple features, but its dimension is high and there is information redundancy, which is not conducive to subsequent analysis and processing. Feature dimension reduction aims to reduce the dimension of data while retaining key information. Feature dimension reduction methods can include principal component analysis (PCA), linear discriminant analysis (LDA), etc. Taking principal component analysis as an example, it converts the original high-dimensional data into a set of unrelated low-dimensional data through linear transformation. These low-dimensional data are called principal components. Each principal component is a linear combination of the original features and retains as much variance information of the original data as possible. By reducing the dimension of the heterogeneous association matrix through principal component analysis, redundant information can be removed, the data structure can be simplified, and the key features that have a greater impact on the cell mechanical phenotype can be retained. Time-frequency joint feature coding is a method for further processing the data after dimensionality reduction. The mechanical behavior of cells has specific characteristics in both time and frequency dimensions, and time-frequency joint feature coding can capture this information at the same time. For example, the reduced-dimensional data is converted from the time domain to the time-frequency domain using methods such as wavelet transform. Wavelet transform can decompose the signal into coefficients of different frequency components at different time points. These coefficients reflect the energy distribution of the signal at different times and frequencies. By encoding these time-frequency coefficients, they are combined into a feature vector, which is the mechanical phenotype descriptor of the target cell. The mechanical phenotype descriptor integrates the information of the cell's mechanical characteristics in time and frequency, more comprehensively describes the cell's mechanical phenotype, and provides a more effective data representation for subsequent analysis and model training.

[0080] S400, using a mechanical phenotype analysis model to process the mechanical phenotype descriptor to obtain a heterogeneity prediction result of the target cell; wherein the mechanical phenotype analysis model is a machine learning model pre-trained using a sample mechanical phenotype descriptor of a sample cell.

[0081] It can be understood that machine learning models have powerful pattern recognition and prediction capabilities. By pre-training the model with sample mechanical phenotype descriptors of a large number of sample cells, the model can learn the potential relationship between mechanical phenotype descriptors and cell heterogeneity. When the mechanical phenotype descriptor of the target cell is input, the model predicts the heterogeneity of the target cell based on its learned pattern. Different types of machine learning models, such as neural networks, support vector machines, etc., can be used to build mechanical phenotype analysis models. Taking a neural network as an example, it consists of multiple neurons and learns patterns in the data by adjusting the connection weights between neurons. During the training process, the mechanical phenotype descriptor of the sample cell is used as input, and the corresponding cell heterogeneity label (such as the degree of differentiation of the cell, functional state, etc.) is used as output. The model continuously adjusts the weights to minimize the error between the predicted result and the actual label. After the training is completed, the model has the ability to predict cell heterogeneity based on the input mechanical phenotype descriptor. For target cells, their mechanical phenotype descriptors are input into the trained model. After internal calculation and processing, the model outputs the prediction results of the target cell heterogeneity, helping researchers understand the differences in mechanical properties between target cells and other cells and their potential biological states, thereby improving the accuracy and generalization of heterogeneity classification.

[0082] Before using the mechanical phenotypic analysis model to process the mechanical phenotypic descriptor to obtain the heterogeneity prediction result of the target cell, the method also includes: S510, obtaining a mechanical phenotype descriptor of the sample cells.

[0083] It can be understood that the mechanical phenotypic descriptors of sample cells are the basic data for training mechanical phenotypic analysis models. The process of obtaining the mechanical phenotypic descriptors of sample cells can be encoded and generated by obtaining information such as the mechanically sensitive areas, local deformation characteristics, and culture environment parameters of the sample cells. For example, when studying the mechanical phenotype of tumor cells, sample cells can include tumor cells from different tumor stages and different tissue sources, as well as normal cells as controls. By obtaining the mechanical phenotypic descriptors of a large number of diverse sample cells, the model can learn a wider range of relationships between cell mechanical phenotypes and heterogeneity, improve the generalization ability of the model, and make it more accurate and reliable when predicting heterogeneity of target cells.

[0084] S520, determining heterogeneity analysis labels of sample cells.

[0085] It can be understood that the heterogeneity analysis label is information used to mark the heterogeneity characteristics of sample cells, which provides a supervision signal for model training. The heterogeneity of sample cells can be reflected in many aspects, such as the degree of cell differentiation, proliferation ability, sensitivity to drugs, disease status, etc. The degree of cell differentiation can be determined by detecting the expression level of cell-specific markers. For example, during the differentiation of neural stem cells, the expression of specific neural marker proteins will change. The expression of these markers can be detected by experimental methods such as immunofluorescence and Western blot to determine the differentiation stage of cells and use them as heterogeneity analysis labels; for tumor cells, heterogeneity analysis labels can be determined based on the grade, stage, and gene mutation of the tumor.

[0086] S530, taking the mechanical phenotype descriptor of the sample cells as input and the corresponding heterogeneity analysis label as expected output, training the initial machine learning model to obtain a trained mechanical phenotype analysis model.

[0087] It can be understood that the mechanical phenotype descriptor of the sample cells is used as the input of the model, and the corresponding heterogeneity analysis label is used as the expected output. The model continuously adjusts the internal parameters (such as the weights of the neural network) to make the model's predicted output as close to the expected output as possible. During the training process, a preset loss function can be used to measure the difference between the model's prediction results and the expected output, such as mean square error (MSE), cross entropy loss, etc. The model calculates the gradient of the loss function to the parameters through the back propagation algorithm, and updates the parameters according to the gradient, so that the loss function gradually decreases. As the training progresses, the model continues to learn the patterns in the sample data, and has a deeper understanding of the relationship between the mechanical phenotype descriptor and the heterogeneity analysis label. After multiple rounds of training, when the model's loss on the training set reaches a certain convergence standard, or the performance on the validation set no longer improves, the model training is considered to be completed, and the trained mechanical phenotype analysis model is obtained. The training data can be shown in the following table: Mechanotype descriptors are feature vectors that integrate mechanosensitive regions, local deformation characteristics, and culture environment parameters, such as: The first three dimensions: the proportion of mechanically sensitive area, the mean value of strain gradient, and the local deformation frequency; The last three dimensions: culture temperature (°C), pH value, and glucose concentration (mM); The heterogeneity label is the degree of differentiation: including poor differentiation, moderate differentiation, and high differentiation.

[0088] The trained model has the ability to predict cell heterogeneity based on the input mechanical phenotype descriptors, and can be used to predict and analyze the heterogeneity of target cells, providing valuable reference for cell biology research.

[0089] Corresponding to the cell mechanical phenotype analysis method based on deep learning in the above embodiment, the embodiment of the present application also provides a cell mechanical phenotype analysis system based on deep learning, and each unit of the system can implement each step of the cell mechanical phenotype analysis method based on deep learning. Figure 3 A structural block diagram of a deep learning-based cell mechanical phenotype analysis system provided in an embodiment of the present application is shown. For ease of explanation, only the parts related to the embodiment of the present application are shown.

[0090] Reference Figure 3 , the deep learning-based cell mechanical phenotyping system includes: An acquisition unit, used to acquire time-series images, mechanical response data and culture environment parameters of target cells; A segmentation unit, configured to perform regional segmentation on the time series image based on the mechanical response data and the culture environment parameters, and determine the mechanically sensitive area on the surface of the target cell and the local deformation characteristics of the subcellular structure; A generating unit, configured to generate a mechanical phenotype descriptor of the target cell according to the mechanically sensitive region, the local deformation characteristics and the culture environment parameters; A result unit is used to process the mechanical phenotype descriptor using a mechanical phenotype analysis model to obtain a heterogeneity prediction result of the target cell; wherein the mechanical phenotype analysis model is a machine learning model pre-trained using a sample mechanical phenotype descriptor of a sample cell.

[0091] It should be noted that the information interaction, execution process, etc. between the above-mentioned systems / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0092] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit module can exist physically separately, or two or more unit modules can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0093] The embodiment of the present application also provides an electronic device, Figure 4 This is a schematic diagram of the structure of an electronic device provided by an embodiment of the present application. Figure 4 As shown, the electronic device 6 of this embodiment includes: at least one processor 60 ( Figure 4 Only one is shown), at least one memory 61 ( Figure 4 Only one is shown in the figure) and a computer program 62 stored in the at least one memory 61 and executable on the at least one processor 60. When the processor 60 executes the computer program 62, the electronic device 6 implements the steps in any of the above-mentioned cell mechanical phenotype analysis method embodiments based on deep learning, or implements the functions of each unit in the above-mentioned system embodiments.

[0094] Exemplarily, the computer program 62 may be divided into one or more units, which are stored in the memory 61 and executed by the processor 60 to complete the present application. The one or more units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program 62 in the electronic device 6.

[0095] The electronic device 6 may be a computing device or terminal device such as a desktop computer, a notebook, a PDA, or a cloud server. The electronic device may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art will appreciate that Figure 4 It is only an example of the electronic device 6 and does not constitute a limitation on the electronic device 6. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, buses, etc.

[0096] The processor 60 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0097] In some embodiments, the memory 61 may be an internal storage unit of the electronic device 6, such as a hard disk or memory of the electronic device 6. In other embodiments, the memory 61 may also be an external storage device of the electronic device 6, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 6. Further, the memory 61 may also include both an internal storage unit and an external storage device of the electronic device 6. The memory 61 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program. The memory 61 may also be used to temporarily store data that has been output or is to be output.

[0098] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0099] An embodiment of the present application provides a computer program product. When the computer program product is executed on an electronic device, the electronic device implements the steps in any of the above method embodiments.

[0100] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to an electronic device, a recording medium, a computer memory, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), an electric carrier signal, a telecommunication signal, and a software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0101] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0102] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0103] In the embodiments provided in the present application, it should be understood that the disclosed cell mechanical phenotype analysis method and device based on deep learning can be implemented in other ways. For example, the cell mechanical phenotype analysis method and device embodiments based on deep learning described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0104] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0105] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A cell mechanical phenotype analysis method based on deep learning, characterized in that: include: Acquire time-series images, mechanical response data, and culture environment parameters of target cells; Based on the mechanical response data and the culture environment parameters, the time series images are segmented to determine the mechanically sensitive areas on the surface of the target cells and the local deformation characteristics of the subcellular structures; generating a mechanical phenotype descriptor of the target cell according to the mechanically sensitive region, the local deformation characteristics and the culture environment parameters; The mechanical phenotype descriptor is processed using a mechanical phenotype analysis model to obtain a heterogeneity prediction result of the target cell; wherein the mechanical phenotype analysis model is a machine learning model pre-trained using a sample mechanical phenotype descriptor of a sample cell.

2. The cell mechanical phenotype analysis method based on deep learning according to claim 1, characterized in that: The method of performing regional segmentation on the time series image based on the mechanical response data and the culture environment parameters to determine the mechanically sensitive area on the surface of the target cell and the local deformation characteristics of the subcellular structure includes: Performing time-varying spatial registration processing on the time-series images to generate a time-varying spatial correlation relationship between the cell deformation field of the target cell and the motion trajectory of the subcellular structure; Based on the time-varying spatial correlation relationship and the mechanical response data, the strain gradient tensor of the target cell surface is calculated to obtain an initial segmentation boundary of the mechanically sensitive area; According to the strain gradient tensor and the culture environment parameters, the initial segmentation boundary is optimized to determine the mechanically sensitive area on the surface of the target cell; The motion trajectory of the subcellular structure is extracted from the time-varying spatial correlation relationship, the curvature change parameter of the subcellular structure is calculated, and the local deformation characteristics of the subcellular structure are determined.

3. The cell mechanical phenotype analysis method based on deep learning according to claim 2, characterized in that: The performing time-varying spatial registration processing on the time-series images to generate the time-varying spatial correlation relationship between the cell deformation field of the target cell and the motion trajectory of the subcellular structure includes: Performing displacement field calculation and feature matching on the time-series images to determine the cell deformation field and subcellular structure motion trajectory of the target cell; Aligning the cell deformation field with the motion trajectory of the subcellular structure to establish a time-varying spatial mapping relationship between the cell deformation field and the subcellular structure; The mapping relationship is subjected to multi-scale strain analysis to generate a time-varying spatial correlation relationship between the cell deformation field of the target cell and the motion trajectory of the subcellular structure.

4. The cell mechanical phenotype analysis method based on deep learning according to claim 3, characterized in that: The step of performing displacement field calculation and feature matching on the time series images to determine the cell deformation field and subcellular structure motion trajectory of the target cell includes: Perform pixel-by-pixel displacement calculation on adjacent time-series images to generate an initial displacement field of the target cell; Identifying subcellular structure feature points from the time-series images, and determining a motion trajectory coordinate sequence of the subcellular structure based on the subcellular structure feature points; Displacement correction is performed based on the initial displacement field and the motion trajectory coordinate sequence of the subcellular structure to determine the cell deformation field and the subcellular structure motion trajectory of the target cell.

5. The cell mechanical phenotype analysis method based on deep learning according to claim 2, characterized in that: The step of calculating the strain gradient tensor of the target cell surface based on the time-varying spatial correlation relationship and the mechanical response data to obtain the initial segmentation boundary of the mechanically sensitive area includes: Based on the time-varying spatial correlation relationship, obtaining the strain tensor component of the target cell surface; Performing spatial derivative operations on the strain tensor components on the surface of the target cell to generate a strain gradient tensor field; Determining a segmentation threshold of a mechanically sensitive area based on the strain gradient tensor field and the mechanical response data; The surface area of ​​the target cell is divided according to the segmentation threshold to obtain an initial segmentation boundary of the mechanically sensitive area.

6. The cell mechanical phenotype analysis method based on deep learning according to claim 5, characterized in that: The step of optimizing the initial segmentation boundary according to the strain gradient tensor and the culture environment parameters to determine the mechanically sensitive area on the surface of the target cell comprises: Dynamically adjusting the segmentation threshold of the initial segmentation boundary based on the culture environment parameters to generate an optimized threshold of a mechanically sensitive area; Performing morphological closing operation on the initial segmentation boundary according to the optimization threshold to generate an optimized boundary of the mechanical sensitive area; The optimized boundary is fused according to the strain gradient tensor to determine the mechanically sensitive area on the surface of the target cell.

7. The cell mechanical phenotype analysis method based on deep learning according to claim 2, characterized in that: The step of extracting the subcellular structure motion trajectory from the time-varying spatial correlation relationship, calculating the curvature change parameter of the subcellular structure, and determining the local deformation characteristics of the subcellular structure includes: Extracting the subcellular structure motion trajectory from the time-varying spatial correlation relationship, and determining the motion trajectory coordinate sequence of the subcellular structure of the target cell; Performing spline interpolation on the motion trajectory coordinate sequence to generate a motion trajectory curve of the subcellular structure, and calculating a trajectory curvature parameter of the subcellular structure according to the motion trajectory curve; A time derivative operation is performed based on the trajectory curvature parameters of the subcellular structure to determine the local deformation characteristics of the subcellular structure.

8. The cell mechanical phenotype analysis method based on deep learning according to claim 1, characterized in that: The step of generating the mechanical phenotype descriptor of the target cell according to the mechanically sensitive area, the local deformation characteristics and the culture environment parameters comprises: Performing feature association according to the mechanically sensitive area, the local deformation feature and the culture environment parameter to generate a heterogeneous association matrix of the target cells; The heterogeneous association matrix is ​​subjected to feature dimensionality reduction, and the reduced heterogeneous association matrix is ​​subjected to time-frequency joint feature encoding to generate a mechanical phenotype descriptor of the target cell.

9. The cell mechanical phenotype analysis method based on deep learning according to claim 1, characterized in that: Before using the mechanical phenotype analysis model to process the mechanical phenotype descriptor to obtain the heterogeneity prediction result of the target cell, the method further includes: Obtain mechanical phenotype descriptors of sample cells; Determining heterogeneity analysis labels of cells in the sample; The initial machine learning model is trained with the mechanical phenotype descriptor of the sample cells as input and the corresponding heterogeneity analysis label as expected output to obtain the trained mechanical phenotype analysis model.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 9 is implemented.

Images