End-to-end cell senescence state prediction method based on nuclear membrane morphology

Abstract

The present application relates to the technical field of edge detection algorithm, and particularly relates to an end-to-end cell aging state prediction method based on nuclear membrane morphology, which comprises the following steps: step one, collecting a fluorescence image of human fibroblasts subjected to immunofluorescence staining through a fluorescence microscope; step two, performing data enhancement processing on nuclear membrane data, amplifying training data, and improving model generalization capability; step three, generating a prediction of nuclear membrane segmentation mask through a deep neural network, and obtaining mask information; and step four, simultaneously inputting the mask information into an ellipticity module, a smoothness module and a concave degree module for multi-angle calculation and evaluation. The present application evaluates nuclear membrane morphology changes from multiple angles to meet the growing demand for quantification of cell aging state in the field.

Description

Technical Field

[0001] This invention relates to the field of edge detection computing technology, specifically to an end-to-end method for predicting cellular senescence state based on nuclear membrane morphology. Background Technology

[0002] Cellular senescence, an irreversible state of cell cycle arrest, has been considered a driving force of individual aging since its initial discovery in the 1960s. Cellular senescence is accompanied by many abnormal biological changes, among which abnormal nuclear membrane morphology is a conserved indicator of aging. Currently, the assessment and detection of nuclear membrane morphology are relatively limited, lacking a single or universally applicable technical assessment indicator, and existing detection methods have significant errors and are easily influenced by subjective factors. Therefore, developing and establishing new nuclear membrane morphology detection indicators using deep learning methods can more efficiently and accurately assess cellular senescence and facilitate comparisons of the effects of different treatments on cellular senescence. In the context of the rapid development of research related to cellular senescence, only by providing comprehensive and accurate cellular senescence assessment indicators can we more effectively guide biological experiments, measure the merits of various protection and restoration methods, quantify the restorative effects on senescent cells, and better provide fundamental support for the vigorous development of the research field. Summary of the Invention

[0003] This invention provides an end-to-end method for predicting cellular senescence based on nuclear membrane morphology, which assesses changes in nuclear membrane morphology from multiple perspectives to meet the growing demand in the field for quantification of cellular senescence status.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an end-to-end method for predicting cellular senescence state based on nuclear membrane morphology, comprising the following steps:

[0005] Step 1: Acquire fluorescence images of human fibroblasts stained with immunofluorescence using a fluorescence microscope;

[0006] Step 2: Perform data augmentation on the nuclear membrane data to expand the training data and improve the model's generalization ability;

[0007] Step 3: Generate predictions of the nuclear membrane segmentation mask using a deep neural network to obtain mask information;

[0008] Step 4: Simultaneously input the mask information into the ellipticity module, smoothness module, and concavity module for multi-angle calculation and evaluation.

[0009] Preferably, in step four, the ellipticity module first constructs a fitted ellipse based on the edge information of the nuclear membrane segmentation mask; then it calculates the shortest distance from each point on the segmentation map to the fitted ellipse and calculates the standard deviation.

[0010] Preferably, in step four, the smoothness module first uses the Douglas-Peucker algorithm to approximate the edge of the nuclear membrane segmentation mask with polygons and use it as the skeleton of the nuclear membrane to eliminate the influence of local jagged edges and make the nuclear membrane edge continuous; then it detects all inflection points on the nuclear membrane skeleton, calculates the curvature to determine the concavity and convexity direction of the inflection points; finally, it calculates the overall degree of wrinkling based on the area enclosed by the concave point and the nuclear membrane convex hull and the maximum concave angle.

[0011] Preferably, in step four, the indentation module first calculates the normal vector of each point based on the edge coordinates of the segmentation mask; then, it uses the Sobel operator to calculate the gradient map of the nuclear membrane data and extracts the nuclear membrane edge information; then, it searches for the second maximum point along the normal from the outer boundary on the gradient map to determine the coordinates of the inner boundary; then, it performs outlier detection on all detected inner boundary points and filters out inaccurate inner boundary points; the area enclosed by all inner boundary points that meet the conditions is the internal region of the cell nucleus after the nuclear membrane is removed; finally, it uses an adaptive threshold method to statistically analyze the indentation signal of the internal region in the gradient map and calculates the ratio to the statistical perimeter.

[0012] The beneficial effects of this invention are as follows: using human skin fibroblast nuclei as experimental subjects, it predicts cell senescence from multiple dimensions of nuclear membrane morphology, comprehensively covering important morphological features of the cell nucleus. Furthermore, it novelly proposes a method for quantifying the indentation information within the cell nucleus, enriching the assessment perspectives of nuclear membrane morphology and cell senescence status. Modular design is adopted in each stage, with smooth internal processes and close cooperation between modules. Moreover, this algorithm has the advantage of strong scalability in its architecture, allowing for easy expansion with other algorithm modules in practical use to achieve more comprehensive evaluation functions. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a diagram of the algorithm architecture of the present invention;

[0015] Figure 2 This demonstrates the semantic segmentation effect of the present invention.

[0016] Figure 3 This is the ellipticity index of the present invention;

[0017] Figure 4 This refers to the smoothness index of the present invention;

[0018] Figure 5 This is the indentation index of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] according to Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 As shown, an end-to-end cellular senescence prediction method based on nuclear membrane morphology includes the following steps:

[0021] Step 1: Acquire fluorescence images of human fibroblasts stained with immunofluorescence using a fluorescence microscope;

[0022] Step 2: Perform data augmentation on the nuclear membrane data to expand the training data and improve the model's generalization ability;

[0023] Step 3: Generate predictions of the nuclear membrane segmentation mask using a deep neural network to obtain mask information;

[0024] Step 4: Simultaneously input the mask information into the ellipticity module, smoothness module, and concavity module for multi-angle calculation and evaluation.

[0025] The above steps combine various artificial intelligence and image processing evaluation algorithms to predict and assess cellular senescence based on different dimensions and features of cell nucleus morphology. Immunofluorescence staining of the nuclear membrane localization protein Lamin A / C is performed, and fluorescence images indicative of nuclear membrane morphology are acquired using a fluorescence microscope. Since the three dimensions of cellular senescence assessment all require nuclear membrane location information as a prerequisite, nuclear membrane segmentation is used as the algorithm front-end. Deep learning technology is then used to predict the edge position of the nuclear membrane, generating a segmentation mask for the cell nucleus. After determining the nuclear membrane coordinates, the back-end visual algorithm is divided into three sub-modules, which calculate and evaluate ellipticity, smoothness, and concavity in parallel based on the mask information.

[0026] exist Figure 1In the algorithm architecture diagram, the deep neural network in the segmentation module is used to predict the nuclear membrane position of cells and segment the cell nucleus from the background, laying the groundwork for the calculation of subsequent indicators. In the ellipticity module, the ellipticity of the cell nucleus indicates the degree of similarity between the cell nucleus morphology and an ellipse: the cell nucleus morphology of young cells is close to an ellipse, while with cell aging, the morphology of the cell nucleus becomes more irregular, resulting in a poor fit to an ellipse. In the smoothness module, the smoothness of the cell nucleus indicates the roundness and smoothness of the nuclear membrane: the cell nucleus of young cells has fewer edge wrinkles, while with the continuous acceleration of cell aging, the wrinkles on the nuclear membrane continuously increase, leading to a continuous decrease in smoothness. In the indentation module, the indentation of the cell nucleus indicates the degree of spatial indentation of the cell nucleus: when we perform fluorescence imaging on the morphology of the cell nucleus and nuclear membrane, we find that fluorescent signals also appear inside the cell nucleus in the 2D nuclear membrane morphology image; these fluorescent signals originate from nuclear membrane indentation. During cell aging, the indentation of the nuclear membrane also continuously increases.

[0027] The outer nuclear membrane has a more coherent and clearer outline than the inner nuclear membrane due to its denser and more uniform pigment granules. Therefore, labeling the outer membrane is significantly easier than labeling the inner membrane. Thus, a supervised deep neural network is used to predict the accurate location information of the nuclear membrane edges. Compared to traditional segmentation algorithms, deep learning algorithms can more accurately segment the location information of the brighter nuclear membrane. This algorithm uses BioNet as the segmentation model, a neural network commonly used for cell segmentation. During annotation, the region with the largest gradient of brightness change in the outer area is selected as the nuclear membrane location for labeling. The dataset is preferably divided into training, validation, and test sets in a 6:2:2 ratio. Testing shows that the segmentation model achieves a 97.88% accuracy rate in segmenting the nuclear membrane location.

[0028] The ellipticity module first constructs a fitted ellipse based on the edge information of the nuclear membrane segmentation mask; then it calculates the shortest distance from each point on the segmentation map to the fitted ellipse and calculates the standard deviation, such as... Figure 3 As shown, the ellipticity module predicts cellular aging by evaluating how closely the cell nucleus fits an ellipse. From a nuclear membrane morphology perspective, the nucleus of young cells is more elliptical in shape, while the morphology of aging cells becomes increasingly irregular with age. Therefore, the normal morphology of the cell nucleus is approximated by calculating the fitted ellipse of the nuclear membrane segmentation mask. For aging cells, the shape of the nucleus will generate more areas that do not coincide with the fitted ellipse. By calculating the shortest distance from each pixel on the nuclear membrane edge to the fitted ellipse, the pixels calculated on the fitted ellipse are considered as mappings to the nuclear membrane edge points, and the standard deviation of all mapped distances is calculated to reflect the degree of aberration. The smaller the ellipticity, the more elliptical the cell nucleus is, and the younger the cell is.

[0029] The smoothness module first uses the Douglas-Peucker algorithm to approximate the edges of the nuclear membrane segmentation mask using polygons, which serve as the framework of the nuclear membrane to eliminate the influence of local jagged edges and make the nuclear membrane edges continuous. Next, it detects all inflection points on the nuclear membrane framework and calculates the curvature to determine the concavity / convexity direction of the inflection points. Finally, it calculates the overall degree of wrinkling based on the area enclosed by the concave points and the nuclear membrane convex hull, as well as the maximum concavity angle. Figure 4 As shown, the smoothness module quantifies the smoothness of the nuclear membrane image. Compared to the smooth, rounded edges of the nuclear membrane in young cells, the nuclear membrane of aging cells bends inward or outward, resulting in many wrinkles at the edges. This fact suggests that the smoothness of the nuclear membrane can be used as an important measure of aging. In the algorithm, the degree of wrinkling on the cell surface is evaluated by statistically analyzing the concavity and convexity of each point on the edge of the nuclear membrane. Semantic segmentation can be viewed as classifying image pixels; therefore, the results obtained by neural networks often exhibit unevenness at the edges. To eliminate the influence of jagged edges on local concavity and convexity, the current submodule first uses the Douglas-Peucker algorithm, segmenting the cell by making secants inside the cell until the minimum distance from all secants to the cell membrane meets our desired threshold. The polygon formed by all secants is used as the cytoskeleton to approximate the boundary of the nuclear membrane, achieving the goal of making the boundary points on the cell nucleus continuous. Then, all inflection points on the cytoskeleton are detected, and the curvature is calculated as discrete points, i.e., the coordinates of the inflection points and the left and right edge points. The concavity / convexity direction of a vertex is determined by the curvature of each inflection point. Finally, the region enclosed by the concave point and the cell's convex hull can be used to measure the degree of cell wrinkling. This region is abstracted as a polygon composed of one or more triangles, and the area of ​​the enclosed region reflects the degree of wrinkling. During cell senescence, the nuclear membrane edge first undergoes a change in depth from shallow to deep, and then a change in angle from small to large. Therefore, wrinkles with different concavity angles and depths may have the same area. We define that, for the same area, the larger the maximum concavity angle, the more severe the wrinkling. Therefore, the quantitative index of this single wrinkle is obtained by multiplying the area and the concavity angle. When multiple wrinkles exist on the nuclear membrane, the formula is: The final outer membrane fold score is calculated in the form of Tn, where Tn represents the nth inward fold on the nuclear membrane.

[0030] The indentation module first calculates the normal vector of each point based on the edge coordinates of the segmentation mask; then, it uses the Sobel operator to calculate the gradient map of the nuclear membrane data to extract the nuclear membrane edge information; next, it searches for the second maximum point along the normal on the gradient map from the outer boundary to determine the coordinates of the inner boundary; then, it performs outlier detection on all detected inner boundary points to filter out inaccurate inner boundary points; the region enclosed by all inner boundary points that meet the conditions is the internal region of the cell nucleus after nuclear membrane removal; finally, it uses an adaptive thresholding method to statistically analyze the indentation signal of the internal region in the gradient map and calculates the ratio to the statistical perimeter, such as... Figure 5 As shown.

[0031] The nuclear membrane indentation module uses immunofluorescence to label nuclear membrane morphology. In the 2D visualization, the spatial indentation of the nuclear membrane causes fluorescence signals within the nucleus, and this fluorescence signal increases with cellular senescence. This nuclear membrane indentation is used as an important indicator of nuclear membrane morphological changes. Compared to other nuclear membrane studies, this method uses a different staining approach. Immunofluorescence staining only stains the nuclear membrane, preserving the visualization of the nucleus. During fluorescence microscopy imaging, the three-dimensional nucleus is presented in a two-dimensional format. In this process, fluorescence signals from different dimensions are reduced to a single space, and the "peaks and valleys" on the nuclear surface are mapped to the cytoplasm. Therefore, the fluorescence signal within the nuclear membrane can also serve as a characterization of the overall degree of nuclear indentation, profoundly reflecting the cell cycle. This module's algorithm effectively reflects the degree of nuclear indentation in three-dimensional space by statistically analyzing high-frequency gradient information within the nucleus.

[0032] In the indentation index module, this algorithm uses the Sobel operator to extract the first-order gradient information of the experimental object from both horizontal and vertical directions. The Sobel operator weights the difference between the gray values ​​of the four neighborhoods (top, bottom, left, and right) of each pixel, reaching an extreme value at the edge to detect the edge. In microscopic imaging, the nuclear membrane edge has a high gradient difference, which can affect signal sampling in the internal region. Therefore, to facilitate the statistical analysis of high-frequency information inside the cell nucleus, it is first necessary to eliminate the influence of fluorescence signals at the nuclear membrane boundary.

[0033] This algorithm aims to statistically analyze internal signals by stripping away the nuclear membrane boundary while preserving the nucleoplasm. To achieve this, a first-order gradient image is constructed using the Sobel operator to extract the edge information of the cell nucleus. Then, for each edge point of the nuclear membrane, a tangent vector is constructed based on the four adjacent points to calculate the normal vector at that point, with the normal direction aligned with the direction within the cell nucleus. Next, the algorithm searches for the brightness maxima in the gradient image from the outside in along the normal coordinates to locate the inner boundary of the membrane. Since the nuclear membrane has a certain thickness, there are drastic gradient changes at both the inner and outer boundaries; therefore, the second brightness maxima found represents the coordinates of the inner membrane. Outlier detection is used during the search to avoid misjudgments caused by blurred edges; distances within twice the standard deviation of the mean thickness (the average distance from all outer membranes to the inner membrane) are considered valid coordinates. Finally, the obtained coordinates are connected, and the enclosed region can be considered the nucleus's internal region after the nuclear membrane has been stripped away. The shape of this internal region largely matches the membrane shape, achieving uniform removal of the nuclear membrane without wasting excessive information from the internal region. After stripping the nuclear membrane, the internal region of the cell nucleus in the gradient image is processed with Gaussian noise, and pixels with larger gradients are selected using an adaptive threshold. These pixels represent the indentation information within the cell nucleus. Since this indentation information is discretely distributed and irregularly shaped, it is difficult to calculate using conventional methods. Therefore, the degree of indentation is approximated by counting the number of pixels; this number is called the indentation length. Cells in the data vary significantly in morphology and size. To standardize these features, the circumference of an equivalent circle representing the area of ​​the internal region of the cell nucleus is used as the statistical perimeter. The ratio of the indentation length to the statistical perimeter is used as the final indentation degree index.

[0034] In summary, this method uses fibroblast nuclei as the experimental subject and predicts cell senescence from multiple dimensions of nuclear membrane morphology, comprehensively covering important morphological features of the cell nucleus. Furthermore, it novelly proposes a method for quantifying the indentation information within the cell nucleus, enriching the assessment perspectives of nuclear membrane morphology and cell senescence status. Modular design is employed in each stage, ensuring smooth workflow within modules and close cooperation between modules. Moreover, this algorithm boasts strong scalability, allowing for easy expansion with other algorithm modules to achieve more comprehensive evaluation functions in practical applications.

[0035] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An end-to-end cellular senescence prediction method based on nuclear membrane morphology, characterized in that... This includes the following steps: Step 1: Acquire fluorescence images of human fibroblasts stained with immunofluorescence using a fluorescence microscope; Step 2: Perform data augmentation on the nuclear membrane data to expand the training data and improve the model's generalization ability; Step 3: Generate predictions of the nuclear membrane segmentation mask using a deep neural network to obtain mask information; Step 4: Simultaneously input the mask information into the ellipticity module, smoothness module, and concavity module for multi-angle calculation and evaluation; The ellipticity module is used to construct a fitted ellipse based on the edge information of the nuclear membrane segmentation mask, and to calculate the standard deviation of the shortest distance from each point on the segmentation map to the fitted ellipse. In step four, the smoothness module first uses the Douglas-Peucker algorithm to approximate the edge of the nuclear membrane segmentation mask with polygons and use it as the skeleton of the nuclear membrane to eliminate the influence of local jagged edges and make the nuclear membrane edge continuous; then it detects all inflection points on the nuclear membrane skeleton, calculates the curvature to determine the concavity and convexity direction of the inflection points; finally, it calculates the overall wrinkle degree based on the area enclosed by the concave point and the nuclear membrane convex hull and the maximum concave angle. The indentation module first calculates the normal vector of each point based on the edge coordinates of the segmentation mask; then it uses the Sobel operator to calculate the gradient map of the nuclear membrane data and extracts the nuclear membrane edge information; next, it searches for the second maximum point along the normal on the gradient map from the outer boundary to determine the coordinates of the inner boundary; then it performs outlier detection on all detected inner boundary points and filters out inaccurate inner boundary points; the area enclosed by all inner boundary points that meet the conditions is the internal region of the cell nucleus after the nuclear membrane is removed; finally, it uses an adaptive threshold method to statistically analyze the indentation signal of the internal region in the gradient map and calculates the ratio of the indentation length to the statistical perimeter.

2. The end-to-end cellular senescence prediction method based on nuclear membrane morphology according to claim 1, characterized in that: In step four, the ellipticity module first constructs a fitted ellipse based on the edge information of the nuclear membrane segmentation mask; then it calculates the shortest distance from each point on the segmentation map to the fitted ellipse and calculates the standard deviation.

Images