Umbilical cord mesenchymal stem cell preparation process monitoring method and system

Abstract

The invention relates to the technical field of image analysis, in particular to an umbilical cord mesenchymal stem cell preparation process monitoring method and system, and the method comprises the following steps: based on an umbilical cord mesenchymal stem cell image, carrying out the image size standardization, noise removal, cell region cutting, and gray value comparison to remove a gray value region lower than a set threshold value; background noise and noisy points are eliminated by applying a filtering method, cell edge information is extracted, and a denoised cell image is generated. According to the method, the residual network and the convolution operation are introduced, the cell morphological change is accurately captured, more accurate proliferation state analysis, cell boundary extraction and cycle classification improvement are provided, misjudgment of the cell state in a traditional method is effectively avoided, image segmentation is optimized through deep learning training, cell cycle division is more accurate, and the method is suitable for large-scale popularization and application. The high efficiency and the accuracy of the whole monitoring process are ensured, and the standardization and the high efficiency of the stem cell preparation process are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and particularly to a method and system for monitoring the preparation process of umbilical cord mesenchymal stem cells. Background Art

[0002] The technical field of image analysis includes technologies for collecting, processing, and analyzing images to extract useful information. The core contents involved in this field include digital image processing, pattern recognition, image segmentation, image enhancement, and feature extraction, etc. Image analysis technology is widely applied in multiple fields such as medical imaging, industrial inspection, machine vision, security monitoring, etc. Through image analysis, different objects, forms, and behaviors can be identified and classified from images, realizing automated data extraction and processing, and thus providing support for various decisions. In the medical field, image analysis is particularly used in aspects such as pathological diagnosis, surgical assistance, disease detection, etc., and has become an important tool for assisting diagnosis and treatment.

[0003] Among them, the method for monitoring the preparation process of umbilical cord mesenchymal stem cells refers to a method for real-time monitoring of the preparation process of umbilical cord mesenchymal stem cells through image analysis technology. This patent theme involves image acquisition and processing during the preparation of umbilical cord mesenchymal stem cells, mainly obtaining image data during the cell culture process through a microscope or other imaging devices, and processing and analyzing these data. This technology solves the problems of how to accurately monitor cell growth, division, state changes, etc. during the stem cell preparation process. Its means include image acquisition, image processing, and analysis, mainly through automated image recognition technology to real-time track and record the cell state, thereby ensuring the standardization and efficiency of stem cell preparation.

[0004] In the prior art, during the preparation process of umbilical cord mesenchymal stem cells, there is a lack of detailed monitoring and real-time feedback on cell morphological changes. Especially, there are often large errors in the recognition of cell proliferation status and morphological changes. The prior art relies on simple image processing methods, which may not be able to effectively handle the dynamic changes of complex cell morphologies, resulting in inaccurate distinction between the cell division and stationary phases. In addition, during the cell cycle classification and image segmentation processes in the prior art, it may be affected by background noise and interference factors, and cannot achieve accurate cycle stage division, thus affecting the accuracy and standardization of the stem cell preparation process. For example, traditional image processing methods may have a blurring phenomenon when processing cell edges, resulting in unclear cell boundary recognition, and thus affecting subsequent analysis and judgment. Summary of the Invention

[0005] The object of the present invention is to solve the deficiencies existing in the prior art, and to propose a method and system for monitoring the preparation process of umbilical cord mesenchymal stem cells.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A method for monitoring the preparation process of umbilical cord mesenchymal stem cells, comprising the following steps:

[0007] S1: Based on the umbilical cord mesenchymal stem cell image, perform image size standardization, noise removal, crop the cell region, execute gray value comparison to remove the gray value region below the set threshold, apply a filtering method to eliminate background noise, remove noise points and extract cell edge information, and generate a denoised cell image;

[0008] S2: Based on the denoised cell image, input the cell image into a residual network, perform weighted sum processing on the output features of each layer through convolution operations, apply a non-linear activation function to adjust the convolution kernel size to match the cell morphology change, calculate the influence of the cell proliferation state on the morphological features, and generate a cell edge feature map;

[0009] S3: Based on the cell edge feature map, perform edge change and gray scale comparison, identify and extract the boundary of umbilical cord mesenchymal stem cells, execute region division and boundary extraction, combine cell morphology monitoring information, perform boundary value comparison analysis, and generate a cell boundary interval value;

[0010] S4: Based on the cell boundary interval value, perform cell cycle stage classification, perform periodic analysis through cell boundary changes, set a standard to divide the mitotic phase and the stationary phase, calculate the cell morphology change within the cycle, and generate a cell cycle stage boundary map;

[0011] S5: Based on the cell cycle stage boundary map, perform deep learning training, perform cycle state subdivision and optimization, adjust the depth of the neural network layer and the learning rate, optimize the image segmentation process, adjust the learning rate and weight of each layer, and obtain an optimized segmentation result map.

[0012] The denoised cell image specifically includes a cell image, a noise removal region, and cell edge information. The cell edge feature map includes convolution features, activation function adjustment, and morphological change influence. The cell boundary interval value includes edge change, gray scale comparison, boundary extraction, and morphological monitoring. The cell cycle stage boundary map includes the mitotic phase, the stationary phase, and cell morphology change. The optimized segmentation result map specifically includes cycle state optimization, network layer adjustment, and segmentation result.

[0013] As a further solution of the present invention, the steps for obtaining the denoised cell image are specifically as follows:

[0014] S101: According to the input umbilical cord mesenchymal stem cell image, adjust the size of all images by a unified scaling ratio, and at the same time, according to the set threshold, compare the gray value of each pixel with the threshold for judgment, filter out the gray value less than the threshold, and generate a standardized image;

[0015] S102: In the standardized image, filter the image using a high-pass or low-pass filter to distinguish different noise frequencies, adjust the filtering intensity using the design parameters of the filter, and at the same time, separate the cell region from other regions through threshold determination to establish a denoised cell region;

[0016] S103: Based on the denoised cell region, locate the edge positions by calculating the gray-scale differences between each pixel in the image and its neighboring pixels, compare the edge intensities using a preset threshold, and use the formula:

[0017]

[0018] Calculate the cell edge information to generate a denoised cell image;

[0019] where E represents the edge information volume, I i represents the gray-scale value of the i-th pixel in the image, α is the edge sharpness coefficient, β is the edge smoothness adjustment coefficient, γ is the detail enhancement coefficient, and N is the total number of pixels in the image.

[0020] As a further solution of the present invention, the steps for obtaining the cell edge feature map are specifically as follows:

[0021] S201: Based on the denoised cell image, input the cell image into a residual network, perform weighted processing on the feature maps of each layer through convolution operations, and at the same time, filter the image multiple times using a filter, weight the output features of the differential layer, adjust the convolution kernel size of the differential layer, and calculate the convolution output of each layer to obtain a convolution result image;

[0022] S202: Apply a non-linear activation function to adjust the convolution result image, adjust the non-linear effect of the output features according to the changes in cell morphology, and obtain a feature map matching the cell morphology by adjusting the convolution kernel size;

[0023] S203: Based on the feature map matching the cell morphology, analyze the changes in cell morphological features under different proliferation states according to the relationship between the proliferation state and cell morphology, calculate the proliferation rate and the amplitude of morphological changes, and adjust the feature map to reflect the changes in the cell proliferation state in combination with the influence of cell proliferation on the image features, using the formula:

[0024]

[0025] Calculate the influence of cell proliferation on morphological features to generate a cell edge feature map;

[0026] where C represents the change value of the morphological feature map, F iIt represents the gray value of the i-th pixel in the image. α is the coefficient of morphological change, β is the reference coefficient of morphological change, γ is the influence coefficient of the proliferation rate, δ is the adjustment factor, and M is the total number of pixels in the image.

[0027] As a further solution of the present invention, the steps for obtaining the cell boundary interval value are specifically as follows:

[0028] S301: Based on the cell edge feature map, extract the gray gradient value of the cell edge change. By comparing the gray gradient value with the target threshold, screen the pixel points that conform to the cell boundary characteristics, and generate a cell edge pixel feature set;

[0029] S302: Call the cell edge pixel feature set, calculate the continuity value of the edge change according to the directionality of the edge change and the gray distribution uniformity index, perform region division, and generate a region boundary characteristic set;

[0030] S303: Call the region boundary characteristic set, through the difference value operation between the boundary characteristic parameters, and use the formula:

[0031]

[0032] Calculate the change intensity of the region boundary, compare and extract the maximum change intensity value, and generate the cell boundary interval value;

[0033] Among them, B i represents the boundary interval value, G j represents the gray value of the j-th boundary pixel point, G j+1 represents the gray value of the adjacent pixel point, D j represents the distance between pixel points, is the average value of all distances, α is the adjustment coefficient of gray change, β is the smoothing parameter of the region boundary characteristic set, and n is the number of boundary pixel points.

[0034] As a further solution of the present invention, the steps for obtaining the cell cycle stage boundary map are specifically as follows:

[0035] S401: Based on the cell boundary interval value, through cell cycle stage classification, perform periodic analysis on the cell boundary change, calculate the difference value of the boundary change within the cell cycle, set the standard division of the mitosis period and the stationary period, and generate the cell cycle stage boundary map;

[0036] S402: Call the cell cycle stage boundary map, combine the time characteristics of the boundary change, calculate the cell morphological change in the differential stage, analyze the change amplitude in multiple stages within the cycle, and generate the cell boundary distribution map in the differential stage within the cycle;

[0037] S403: In the cell boundary distribution map in the differential stage within the cycle, use the formula:

[0038]

[0039] Calculate the boundary change intensity of the multi - cell cycle stage, optimize the division of multiple stages within the cycle, and generate a cell cycle stage - divided boundary map;

[0040] Among them, C s represents the boundary change intensity of the cycle stage, P k represents the boundary value of the k - th cell stage, Y k represents the boundary position of the k - th stage, Y k+1 represents the boundary position of the adjacent stage, is the mean value of the boundary positions, β is the adjustment coefficient, γ t is the time adjustment coefficient, and m is the total number of stages within the cell cycle.

[0041] As a further solution of the present invention, the specific steps for obtaining the optimized segmentation result map are as follows:

[0042] S501: Based on the cell cycle stage - divided boundary map, perform cycle state subdivision and optimization through deep - learning training. Combine image data and a neural network model, adjust the depth of the neural network layer and the learning rate, and generate an optimized neural network model;

[0043] S502: Call the optimized neural network model, adjust the learning rate and weight of each layer, perform backpropagation and optimize the image segmentation process, calculate the gradient value of the weight of each layer, and generate an optimized segmentation strategy;

[0044] S503: In the application of the optimized segmentation strategy, use the formula:

[0045]

[0046] Calculate the error function of each neural network layer and perform parameter adjustment to generate an optimized segmentation result map;

[0047] Among them, L represents the loss function value, Z i represents the true value of the i - th image pixel, represents the predicted value, λ is the regularization coefficient, α j is the learning rate adjustment coefficient of layer j, W j represents the weight value of the j - th layer, n is the total number of pixel points, and m is the number of network layers.

[0048] Umbilical cord mesenchymal stem cell preparation process monitoring system. The umbilical cord mesenchymal stem cell preparation process monitoring system is used to execute the above - mentioned umbilical cord mesenchymal stem cell preparation process monitoring method. The system includes:

[0049] Based on the umbilical cord mesenchymal stem cell image, the image processing module standardizes the image size, adjusts the aspect ratio of the image, performs noise removal to remove the background noise of the image, crops the cell region, removes the gray value regions below the set threshold, eliminates the noise points in the image through gray value comparison operations, extracts the edge information of the cells, and generates a denoised cell image;

[0050] Based on the denoised cell image, the feature extraction module inputs the image into the residual network, processes the output features through convolution operations, adjusts the convolution kernel size to match the morphological changes of the cells, calculates the influence of the proliferation state on the cell morphology, extracts the features of the cell edges, and generates a cell edge feature map;

[0051] Based on the cell edge feature map, the boundary recognition module conducts edge change and gray value comparison, identifies the boundary region between the cells and the background, extracts the edge positions of the cells, performs region division and boundary extraction, combines the cell morphological information, and conducts boundary value comparison analysis to obtain the cell boundary interval value;

[0052] Based on the cell boundary interval value, the cycle analysis module classifies the cell cycle stages, analyzes the periodicity through cell boundary changes, determines whether the cells are in the mitotic or quiescent phase, calculates the morphological changes of the cells within the cycle, and obtains a cell cycle stage boundary map;

[0053] Based on the cell cycle stage boundary map, the deep learning optimization module conducts deep learning training, optimizes the hierarchical depth and learning rate of the neural network, adjusts the learning rate and weights of each layer, optimizes the image segmentation process, and generates an optimized cell segmentation result map.

[0054] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0055] In the present invention, image standardization and denoising processing provide more reliable basic data for subsequent analysis, reducing the influence of external interference on the analysis results. The introduction of the residual network accurately models the dynamic changes of cell morphology through deep convolution and nonlinear activation functions, ensuring that the morphological changes of cells at different stages can be captured in a timely and accurate manner. The accurate classification of cell cycle stages not only improves the monitoring fineness through periodic analysis but also avoids misdiagnosis or missed diagnosis problems caused by inaccurate cycle classification. Finally, the optimization of deep learning training improves the accuracy of image segmentation, making the judgment of cell states more scientific and providing a more efficient and accurate monitoring means for the stem cell preparation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 is a schematic diagram of the working process of the present invention;

[0057] Figure 2 It is a flowchart of the acquisition steps for denoising cell images of the present invention;

[0058] Figure 3 It is a flowchart of the acquisition steps for the cell edge feature map of the present invention;

[0059] Figure 4 It is a flowchart of the acquisition steps for the cell boundary interval value of the present invention;

[0060] Figure 5 It is a flowchart of the acquisition steps for the cell cycle stage boundary map of the present invention;

[0061] Figure 6 It is a flowchart of the acquisition steps for the optimized segmentation result map of the present invention. Detailed implementation manners

[0062] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0063] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In addition, in the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0064] Embodiment 1

[0065] Please refer to Figure 1 , the present invention provides a technical solution: a method for monitoring the preparation process of umbilical cord mesenchymal stem cells, including the following steps:

[0066] S1: Based on the umbilical cord mesenchymal stem cell image, perform image size standardization, noise removal, crop the cell region, execute gray value comparison to remove the gray value region below the set threshold, apply a filtering method to eliminate background noise, remove noise points and extract cell edge information, and generate a denoised cell image;

[0067] S2: Based on the denoised cell image, input the cell image into the residual network, process the output features of each layer through convolution operations for weighted summation, apply a non-linear activation function to adjust the convolution kernel size to match the cell morphological changes, and calculate the influence of the cell proliferation state on the morphological features to generate a cell edge feature map;

[0068] S3: Based on the cell edge feature map, conduct edge change and grayscale contrast, identify and extract the boundaries of umbilical cord mesenchymal stem cells, perform region division and boundary extraction, combine cell morphology monitoring information, conduct boundary value comparison and analysis, and generate cell boundary interval values;

[0069] S4: Based on the cell boundary interval values, conduct cell cycle stage classification, perform periodic analysis through cell boundary changes, set standards to divide the mitotic phase and the stationary phase, calculate the cell morphology changes within the cycle, and generate a cell cycle stage boundary map;

[0070] S5: Based on the cell cycle stage boundary map, perform deep learning training, conduct cycle state refinement and optimization, adjust the depth of the neural network layers and the learning rate, optimize the image segmentation process, adjust the learning rate and weights of each layer, and obtain an optimized segmentation result map.

[0071] The denoised cell image specifically includes a cell image, a noise removal area, and cell edge information. The cell edge feature map includes convolution features, activation function adjustment, and morphological change effects. The cell boundary interval values include edge change, grayscale contrast, boundary extraction, and morphology monitoring. The cell cycle stage boundary map includes the mitotic phase, the stationary phase, and cell morphology changes. The optimized segmentation result map specifically includes cycle state optimization, network layer adjustment, and segmentation results.

[0072] Please refer to Figure 2 , and the specific steps for obtaining the denoised cell image are as follows:

[0073] S101: According to the input umbilical cord mesenchymal stem cell images, adjust the size of all images by a unified scaling ratio. At the same time, based on the set threshold, compare the grayscale value of each pixel with the threshold for judgment, filter out the grayscale values less than the threshold, and generate a standardized image;

[0074] First, all images need to be resized by a unified scaling ratio, which is calculated from the ratio of the size of the input image to the required output size. The specific formula is: Scaling ratio = Output size / Input size. In practical applications, the size of the input image can be arbitrary, and the output size can be set according to experimental requirements. For example, the output is set to 512×512 pixels. Therefore, the scaling ratio is obtained by calculating the ratio of the sizes of the input and output images. This step is an important process to ensure that all input images reach a unified size. Next, a comparison is made based on the set gray-scale threshold. The gray-scale value of each pixel in the image is judged against the threshold, and those gray-scale values less than the threshold are removed. The purpose of this process is to remove irrelevant background noise by removing pixels with lower gray-scale values and enhance the saliency of the cell region. The comparison of the gray-scale value with the threshold can be completed through simple numerical calculations. Set a threshold T. If the pixel gray-scale value I is less than T, then the pixel value is set to 0, indicating that the pixel is filtered out; otherwise, the pixel value is retained. In actual operation, the selected threshold T is determined by statistically analyzing the gray-scale distribution of the pixels in the image. For example, an appropriate threshold range is set according to the average or median of the gray-scale values of all pixels in the input image to ensure that most of the effective cell regions are retained while removing irrelevant regions with lower gray-scale values. When analyzing the image, if the threshold T is set to 80, then the pixels in the image with gray-scale values less than 80 will be filtered out, and the pixel region greater than or equal to 80 will be retained, thus generating a standardized image. Through this process, effective filtering of the image can be achieved, and a standardized image with background noise removed can be obtained.

[0075] S102: In the standardized image, use a high-pass or low-pass filter to filter the image, distinguish different noise frequencies, adjust the filtering intensity using the design parameters of the filter, and at the same time, through threshold determination, separate the cell region from other regions to establish a denoised cell region;

[0076] Filter the image. Process the image using a high-pass or low-pass filter. Which filter to choose specifically depends on the characteristics of the noise frequency in the image. If there is low-frequency noise in the image, a low-pass filter will be selected to remove the noise; conversely, if the noise is a high-frequency component, a high-pass filter is used. The selection and design of the filter need to be adjusted according to the characteristics of the image noise. For example, a low-pass filter can smooth the high-frequency noise in the image, while a high-pass filter can highlight the edge features in the image. The design parameters of the filter, such as the cut-off frequency, need to be obtained through spectral analysis tools based on the noise characteristics. Spectral analysis can help identify the main frequency components in the image, and then determine the appropriate filter parameters. This process can be analyzed and filtered using image processing software such as MATLAB or OpenCV in Python during actual operation. During the filtering process, the intensity of filtering is controlled by selecting appropriate design parameters (such as the cut-off frequency or attenuation coefficient of the filter). In practical applications, when setting the parameters of the filter, a series of experiments will be carried out to verify the optimal filtering intensity. For example, assume that in image processing, we select a low-pass filter with a cut-off frequency set to 0.3, which means the filter can remove signals with frequencies higher than 0.3 and retain the low-frequency information. In this way, the high-frequency noise unrelated to cells in the image can be effectively removed, and the main information of the cell area can be retained. After the filtering operation is completed, a threshold is set to determine which areas belong to the cell area. This process is completed by comparing the pixels in the image with the set threshold. If the pixel value is higher than the threshold, the area is regarded as the cell area; otherwise, it is excluded. The generated denoised cell area image is obtained by combining a high-pass or low-pass filter and threshold determination.

[0077] S103: Based on the denoised cell area, locate the edge positions by calculating the gray-scale differences between each pixel in the image and its neighboring pixels, compare the edge intensities using a preset threshold, and adopt the formula:

[0078]

[0079] Calculate the cell edge information to generate a denoised cell image;

[0080] where, E represents the edge information volume, I i represents the gray-scale value of the i-th pixel in the image, α is the edge sharpness coefficient, β is the edge smoothness adjustment coefficient, γ is the detail enhancement coefficient, and N is the total number of pixels in the image.

[0081] Formula:

[0082]

[0083] The advantage of the formula is that by introducing the gray - value difference, edge sharpness coefficient, smoothness adjustment coefficient, and detail enhancement coefficient, it can perform fine calculations on the cell edges in the image. In particular, it can remove the influence of noise, retain the clear edges of the cell region, and effectively improve the quality and details of the image.

[0084] Detailed explanation of the formula and the derivation process of formula calculation:

[0085] E represents the edge information quantity, indicating the sharpness and intensity of the edges in the image.

[0086] I i represents the gray - value of the i - th pixel in the image.

[0087] α is the edge sharpness coefficient, used to adjust the edge sharpness, obtained from experimental data, and the reasonable range is α = 1.0 to 2.0.

[0088] β is the edge smoothness adjustment coefficient, used to smooth the transition of the edge region, obtained through experimental analysis, and the reasonable value range is β = 0.1 to 1.0.

[0089] γ is the detail enhancement coefficient, used to enhance the details in the image, and its value is set as γ = 1.0 to 2.0, used to adjust the degree of detail prominence in the image.

[0090] N represents the total number of pixels in the image, used to represent the overall size of the image.

[0091] Assume that in actual operation, the gray - value and edge features of the image have been obtained through the aforementioned standardization and filtering processes, and the parameters are set as follows:

[0092] α = 1.5 (edge sharpness coefficient);

[0093] β = 0.5 (edge smoothness adjustment coefficient);

[0094] γ = 1.2 (detail enhancement coefficient);

[0095] Assume that the gray - value difference between neighboring pixels in the image is |I i -I i-1 | = 20.

[0096] The total number of pixels N = 512×512 = 262144 (assuming the image is of size 512×512);

[0097] According to the formula for calculation:

[0098]

[0099] Assume that the gray - value difference of each pixel is 20, and the formula calculation is as follows:

[0100]

[0101] Calculated as:

[0102]

[0103] The calculation result of the edge information amount E is a fixed value. This result indicates that the clarity and details of the edge have been optimized through the adjustment of the formula, and the edge of the image cell region has been captured more precisely, and the influence of noise on the image has been effectively suppressed.

[0104] Please refer to Figure 3 , and the specific steps for obtaining the cell edge feature map are as follows:

[0105] S201: Based on the denoised cell image, input the cell image into the residual network, perform weighted processing on the feature map of each layer through convolution operations, filter the image multiple times through filters, output the features of the weighted differentiation layer, adjust the convolution kernel size of the differentiation layer, and calculate the convolution output of each layer to obtain the convolution result image;

[0106] First, input the denoised cell image into the residual network. The size of the input image is a fixed value, such as 512x512 pixels. After preprocessing to remove noise, the image will be input into the residual network as input. The residual network is processed through multiple convolutional layers, and each layer of convolution operation extracts the features of the image. The key to the convolution operation is to use the convolution kernel to perform weighted processing on the image and extract local features. For example, a 3x3 convolution kernel can be used to extract the edge features in the image, while a 5x5 convolution kernel can capture a larger range of image information. The combination of different convolution kernels can enhance the network's ability to process different details in the cell image. After the convolution operation, through weighted adjustment, the network performs weighted fusion on the feature maps of the convolution results of each layer to highlight important features and suppress unnecessary background information. During this process, the size of the convolution kernel can be adjusted according to the characteristics of the cell image. Smaller cell morphologies may require the use of smaller convolution kernels, while larger cell morphologies are suitable for using larger convolution kernels. Through multiple filtering operations, the local features of the image are gradually enhanced, and the output features of the differentiation layer are further processed to ensure the extraction of the key information of the cell morphology.

[0107] S202: Apply a non-linear activation function to adjust the convolution result image, adjust the non-linear effect of the output features according to the changes in the cell morphology and the differential morphology of the cells, and obtain a feature map that matches the cell morphology by adjusting the convolution kernel size;

[0108] Non-linear activation functions are used to introduce non-linear effects into image features to enhance the expressive power of neural networks. Commonly used non-linear activation functions include ReLU, Sigmoid, Tanh, etc. Among them, ReLU is selected because of its high computational efficiency and its ability to effectively alleviate the vanishing gradient problem. After the feature map output by the convolutional layer is processed by the activation function, the features of the image will be further extracted and enhanced. In cell image analysis, the morphological features of cells (such as size, shape, etc.) will affect the output of the convolutional layer. According to the changes in cell morphology, the effects of non-linear activation functions will vary. For example, for smaller cells, the cell morphology changes less, so the non-linear effect is weaker; while for larger cells, the morphological changes are more obvious, so the effect of non-linear activation is more prominent. According to the differentiated cell morphologies, the size of the convolutional kernel will be further adjusted. The process of adjusting the convolutional kernel size is to analyze different cell morphologies, select an appropriate convolutional kernel size to adapt to the features of different cells. Larger cell morphologies may require larger convolutional kernels to extract detailed information inside the cells, while smaller cells may extract the contour features of the cells through smaller convolutional kernels. This adjustment process relies on experimental data and the feature analysis of cell images, testing with different combinations of convolutional kernels to find the most suitable size configuration. In this way, the network can obtain a feature map that matches the cell morphology, further improving the accuracy and effect of cell image analysis.

[0109] S203: Based on the feature map that matches the cell morphology, according to the relationship between the proliferation state and the cell morphology, analyze the changes in the morphological features of cells in different differentiated proliferation states. By calculating the proliferation rate and the amplitude of morphological changes, combined with the influence of cell proliferation on image features, adjust the feature map to reflect the changes in the cell proliferation state, using the formula:

[0110]

[0111] Calculate the influence of cell proliferation on morphological features to generate a cell edge feature map;

[0112] where C represents the change value of the morphological feature map, F i represents the gray value of the i-th pixel in the image, α is the coefficient of morphological change, β is the reference coefficient of morphological change, γ is the influence coefficient of the proliferation rate, δ is the adjustment factor, and M is the total number of pixels in the image.

[0113] Formula:

[0114]

[0115] The advantage of the formula is that by introducing multiple parameters such as the morphological change coefficient, reference coefficient, influence coefficient of proliferation rate, and adjustment factor, it can accurately quantify the impact of cell proliferation on morphological characteristics. Combining with the changes in the morphological feature map can improve the accuracy of cell proliferation state analysis.

[0116] Detailed explanation of the formula and the derivation process of formula calculation:

[0117] C represents the change value of the morphological feature map.

[0118] F i represents the grayscale value of the i-th pixel in the image, F i α represents the morphological change grayscale value of the i-th pixel, represents the reference coefficient grayscale value of the (i - 1)-th pixel.

[0119] α is the coefficient of morphological change, β is the reference coefficient of morphological change, γ is the influence coefficient of proliferation rate, and δ is the adjustment factor.

[0120] M is the total number of pixels in the image. The formula calculates the difference between the grayscale change of each pixel and the grayscale reference value of its previous pixel, and adjusts this difference to reflect the impact of proliferation rate and morphological change on the image features. To obtain the actual parameter values, measurements need to be based on experimental data:

[0121] F i and F i-1 values are obtained through image acquisition and pixel grayscale analysis. Use image processing software or programming tools (such as OpenCV) to grayscale the cell image and extract the grayscale value of each pixel.

[0122] The morphological change coefficient α can be obtained by analyzing the relationship between cell morphological changes and proliferation state. For example, by comparing the changes in cell images at different time points and using statistical methods (such as the least squares method) for fitting.

[0123] The influence coefficient γ of proliferation rate and the adjustment factor δ need to be calculated by combining proliferation experimental data and using the relationship between cell division rate and image feature changes. The proliferation rate can be obtained through time series analysis methods. For example, record the cell division interval and calculate the average proliferation rate. Assume the following specific values are obtained through experiments:

[0124] F i = 120, F i-1 = 100 (pixel grayscale value);

[0125] α = 1.2, β = 0.8, γ = 0.5, δ = 2;

[0126] The total number of pixels in the image M = 1000;

[0127] Calculation and derivation:

[0128]

[0129] 120 1.2 ≈177.83, 100 0.8 ≈63.10, |177.83 - 63.10| = 114.73, |120| 0.5 = 10.95, substituting into the formula gives:

[0130]

[0131] C ≈ 1000 × 58.97 = 58970;

[0132] This result indicates that the change value of the cell morphological feature map is 58970, reflecting the significance of the cell proliferation state and morphological changes.

[0133] Please refer to Figure 4 , and the specific steps for obtaining the cell boundary interval value are as follows:

[0134] S301: Based on the cell edge feature map, extract the gray gradient value of the cell edge change. By comparing the gray gradient value with the target threshold, screen the pixel points that meet the cell boundary characteristics, and generate a cell edge pixel feature set;

[0135] Based on the cell edge feature map, extract the gray gradient value of the edge change by analyzing the gray change characteristics in the image. Call the gray distribution information of specific pixel points in the image, compare the undulating range of the gray change with the set threshold for zoning, and for each region, screen out the pixel points whose gray change gradient exceeds the threshold. Normalize the screened pixel points according to the gray gradient value, extract the average gray distribution of the overall edge and the distribution of local outliers through statistical analysis, and combine the above processing results to form a set of pixel points of the gray change of the cell boundary and mark and store the pixel points to generate a cell edge pixel feature set.

[0136] S302: Call the cell edge pixel feature set, calculate the continuity value of the edge change according to the directionality of the edge change and the gray distribution uniformity index, perform regional division, and generate a regional boundary characteristic set;

[0137] Call the set of pixel features at the cell edge, analyze the directional features of the gray-scale change in the set of pixel points, use a differential operator to calculate the trend of gray-scale change between pixel points, extract the main boundary direction by combining the directional characteristics of the gray-scale change, analyze the uniformity of the gray-scale change in the boundary direction, divide the image by combining the mean value of the gray-scale distribution and the directional characteristics, calculate the continuity of the boundary characteristics of each divided region, associate the change intensity of the boundary pixel points with the gray-scale distribution, form boundary parameters with directional and regional characteristics, store the boundary direction and gray-scale change characteristic information, and generate a set of regional boundary characteristics.

[0138] S303: Call the set of regional boundary characteristics, through the difference value operation between boundary characteristic parameters, use the formula:

[0139]

[0140] Calculate the change intensity of the regional boundary, compare and extract the maximum change intensity value, and generate the cell boundary interval value;

[0141] Among them, B i represents the boundary interval value, G j represents the gray-scale value of the j-th boundary pixel point, G j+1 represents the gray-scale value of the adjacent pixel point, D j represents the distance between pixel points, is the average value of all distances, α is the adjustment coefficient of the gray-scale change, β is the smoothing parameter of the set of regional boundary characteristics, and n is the number of boundary pixel points.

[0142] Formula:

[0143]

[0144] The advantage of the formula is that by introducing the cumulative amount of the boundary gray-scale difference and the standard deviation of the pixel point distance, combined with the smoothing parameter and the adjustment coefficient, the sensitivity and robustness to the boundary change intensity are enhanced.

[0145] Detailed explanation of the formula and the derivation process of the formula calculation:

[0146] 1. Set the number of pixel points in the regional boundary as n = 5, and the gray-scale value sequence as G1 = 50, G2 =

[0147] 60, G3 = 55, G4 = 70, G5 = 65, and the distance sequence as D1 = 2, D2 = 3, D3 = 2.5, D4 =

[0148] 3.5, D5 = 2.8.

[0149] 2. Calculate the cumulative amount of the absolute value of the gray-scale difference:

[0150]

[0151] 3. Calculate the distance standard deviation:

[0152]

[0153] 4. Substitute the adjustment coefficient α = 1.2 and the smoothing parameter β = 0.8 for calculation:

[0154]

[0155] The result shows that the change intensity value of the regional boundary is 6.19, which is related to the gray gradient value and the pixel distance characteristics. This result is used to determine the most significant cell boundaries in the region and generate the cell boundary interval values.

[0156] Please refer to Figure 5 , and the specific steps for obtaining the cell cycle stage boundary map are as follows:

[0157] S401: Based on the cell boundary interval values, through cell cycle stage classification, perform a periodic analysis of the cell boundary changes, calculate the difference values of the boundary changes within the cell cycle, set the standard division between the mitosis stage and the quiescent stage, and generate the cell cycle stage boundary map;

[0158] The cell boundary interval values are obtained by analyzing the edge features in the cell image. By calculating the pixel gray values and the edge change conditions in the image, the boundary change values are generated and the region is divided. First, by selecting the region of interest and preprocessing the cell image, such as removing noise and performing smoothing operations, and then using an edge detection algorithm such as the Canny algorithm to identify the cell boundaries. Through the gray value gradient change, the boundary information of the cells is obtained. Then, the regions in the image are divided, and by calculating the entropy value of each interval, the cell boundary values of each interval are finally obtained, and the cell boundary intervals are further calibrated. At this time, the division of the boundary values provides an important data basis for the further analysis of the cell cycle.

[0159] S402: Call the cell cycle stage boundary map, combine the time characteristics of the boundary changes, calculate the cell morphological changes in the differential stages, analyze the change amplitudes in multiple stages within the cycle, and generate the cell boundary distribution map of the differential stages within the cycle;

[0160] The cell cycle stage boundary map reflects the morphological changes of cells by analyzing the boundary changes of cells in different cycle stages. First, periodic analysis is carried out using the cell boundary interval values, and combined with the image time series data, the boundary changes of cells in different cycle stages are extracted. To ensure the accuracy of cycle division, time series analysis methods such as sliding window analysis and weighted average method are adopted to further calculate the boundary values and morphological changes of cells in different stages. The changes in each stage can be quantified by analyzing features such as the aspect ratio, area, and boundary contour of cells. Through these data, the boundary map of the cell cycle stage is finally generated, which can visualize the morphological changes of cells in each stage, thus supporting periodic analysis and other biological research.

[0161] S403: In the cell boundary distribution map of the differential stage within the cycle, the formula:

[0162]

[0163] Calculate the boundary change intensity of multiple cell cycle stages, optimize the division of multiple stages within the cycle, and generate the cell cycle stage boundary map;

[0164] Among them, C s represents the boundary change intensity of the cycle stage, P k represents the boundary value of the k-th cell stage, Y k represents the boundary position of the k-th stage, Y k+1 represents the boundary position of the adjacent stage, is the mean value of the boundary positions, β is the adjustment coefficient, γ t is the time adjustment coefficient, and m is the total number of stages within the cell cycle.

[0165] Formula:

[0166]

[0167] The benefit of the formula is that by introducing the boundary change intensity (P k ), the time adjustment coefficient (γ t ) and the mean value of the boundary positions , it can more accurately identify the intensity of the boundary change of the cell cycle stage and optimize the division of the cell cycle stage.

[0168] Detailed explanation of the formula and the derivation process of the formula calculation:

[0169] C in the formula s represents the boundary change intensity of the cycle stage, reflecting the amplitude of the boundary change in each stage within the cell cycle, P k represents the boundary value of the k-th cell cycle stage, Y k represents the boundary position of the k-th stage, Yk+1 Indicates the boundary position of adjacent phases, which is the mean of the boundary positions of all phases within the cell cycle, γ t where γ is the time adjustment coefficient, β is the adjustment coefficient, and m is the total number of phases within the cell cycle.

[0170] Assume that there are 5 phases within the cell cycle, i.e., m = 5, and the boundary positions Y1, Y2, Y3, Y4, Y5 of each phase are: 0.5, 0.6, 0.7, 0.8, 0.9 respectively, and the boundary values P1, P2, P3, P4, P5 are: 1.0, 1.2, 1.5, 1.3, 1.1 respectively, and the time adjustment coefficient γ t = 0.02, and the adjustment coefficient β = 0.1.

[0171] First, calculate the absolute value of the change in boundary position:

[0172] |Y1 - Y2| = |0.5 - 0.6| = 0.1, |Y2 - Y3| = |0.6 - 0.7| = 0.1, |Y3 - Y4| = |0.7 - 0.8| = 0.1, |Y4 - Y5| = |0.8 - 0.9| = 0.1;

[0173] Then, calculate the weighted sum of each phase:

[0174]

[0175] Calculate the mean of the boundary positions:

[0176]

[0177] Next, calculate the difference between each boundary and the mean:

[0178]

[0179] Finally, substitute into the formula for calculation:

[0180]

[0181] The result shows that the intensity of the change in the cycle phase boundary is 1.525, indicating that the amplitude of the boundary change of the cell within this cycle is moderate, which can help further optimize the cycle phase division and support the accuracy of cell cycle analysis.

[0182] Please refer to Figure 6 for the specific steps to obtain the optimized segmentation result graph:

[0183] S501: Based on the cell cycle phase boundary graph, perform cycle state subdivision and optimization through deep learning training. Combine image data and a neural network model, adjust the depth of the neural network layer and the learning rate, and generate an optimized neural network model;

[0184] When performing deep learning training and cycle state subdivision based on the cell cycle stage boundary map, first preprocess the input image data. The image size is uniformly adjusted to the input size required by the network. Usually, gray normalization is performed on the cell images to ensure that the pixel values of the input data are within a certain range, remove noise, and enhance the image quality. Then, input the preprocessed image data into the initial layer of the neural network. The convolutional layer (ConvLayer) is used to extract image features. Each layer in the network is responsible for extracting different levels of features, such as the edges and shapes of cells. During the training process, the labeled cell cycle stage data is used as the training set. Through iterative optimization, the neural network can learn how to identify and classify different cell cycle stages from the images. To improve the network training efficiency, the batch normalization method is usually adopted to accelerate the convergence speed of the network and reduce the risk of overfitting at the same time. After the training is completed, the validation set is used for validation, and the hyperparameters of the model, such as the learning rate (LearningRate) and the weight initialization method, are adjusted to improve the generalization ability and accuracy of the model. After training, the neural network can generate an optimized neural network model and is ready to enter the next stage of cell cycle classification and image segmentation tasks.

[0185] S502: Call the optimized neural network model, adjust the learning rate and weight of each layer, perform backpropagation and optimize the image segmentation process, calculate the gradient value of the weight of each layer, and generate an optimized segmentation strategy;

[0186] Call the optimized neural network model. By adjusting the learning rate and weight of each layer, first select an appropriate learning rate adjustment strategy according to the performance of the current model on the validation set. If the current learning rate is relatively large, it may cause the model to oscillate or fail to converge. At this time, the learning rate needs to be reduced; if the learning rate is too small, the training speed will be too slow and the model cannot be effectively optimized. Then, the weight values of each layer of the network are updated according to the backpropagation algorithm (Backpropagation). Backpropagation calculates the error gradient of each layer, reversely transmits the error from the output layer to the input layer, and calculates the gradient value of each layer, and updates the weight of each layer accordingly. During this process, it is necessary to ensure that the adjustment of the learning rate makes the weight of each layer not have problems such as gradient explosion or gradient disappearance during the optimization process. Usually, the momentum method is combined to further accelerate the training. As the neural network training progresses, the weight values of different layers are continuously adjusted after each iteration, further improving the effect of image segmentation. Each training result is used for validation and fine-tuning to improve the accuracy of the model until the performance of the model reaches the best, and an optimized segmentation strategy is generated for the cell cycle image segmentation task in the subsequent stage.

[0187] S503: In the application of the optimized segmentation strategy, the formula:

[0188]

[0189] is used to calculate the error function of each neural network layer, adjust the parameters, and generate an optimized segmentation result graph;

[0190] where L represents the value of the loss function, Z i represents the true value of the i-th image pixel, represents the predicted value, λ is the regularization coefficient, α j is the learning rate adjustment coefficient of layer j, W j represents the weight value of the j-th layer, n is the total number of pixel points, and m is the number of network layers.

[0191] In the application of the optimized segmentation strategy, the formula:

[0192]

[0193] The benefits of the formula are as follows: by introducing a regularization term, it controls the scale of the network weights, avoids overfitting, and when updating the weights through backpropagation, it considers the impact of the weight value on the loss function. Detailed explanation of the formula and the derivation process of the formula calculation:

[0194] 1. L represents the value of the loss function, and by calculating the difference between the predicted value and the true value (i.e., the sum of squared errors), the accuracy of the network output is evaluated;

[0195] 2. Z i represents the true value of the i-th image pixel, is the pixel value predicted by the model, is the prediction error of the i-th pixel, and the sum of the squares is taken to obtain the total error.

[0196] 3. λ is the regularization coefficient, which is used to balance the weights between the error term and the weight regularization term.

[0197] 4. α j is the learning rate adjustment coefficient of layer j, which controls the learning speed of each layer, W j represents the weight value of the j-th layer, represents the squared term of the weight, which reflects the contribution of the weight to the loss function.

[0198] 5. is the regularization of the weight term. By adding this term, the model avoids excessive weights during update and avoids overfitting problems.

[0199] Specific calculation example:

[0200] Suppose there are 3 pixel points, namely Z1 = 3, Z2 = 4, Z3 = 5, and the corresponding predicted values are And the regularization coefficient λ = 0.1, the learning rate adjustment coefficients α1 = 0.5 and α2 = 0.3, the weight value of the first layer W1 = 0.2 and the weight value of the second layer W2 = 0.1. We can calculate the value of the loss function.

[0201] Calculate the sum of squared errors part:

[0202]

[0203] So the sum of squared errors part is:

[0204]

[0205] Calculate the regularization part:

[0206]

[0207] So the regularization part is:

[0208]

[0209] Combined loss function:

[0210] L = 0.3 + 0.1·0.023 = 0.3 + 0.0023 = 0.3023;

[0211] This result shows that the value of the loss function is 0.3023, indicating that under the influence of the error of the current model and the regularization term, the fitting effect of the current model is relatively ideal, but the model weights still need to be further optimized.

[0212] Umbilical cord mesenchymal stem cell preparation process monitoring system. The umbilical cord mesenchymal stem cell preparation process monitoring system is used to execute the above-mentioned umbilical cord mesenchymal stem cell preparation process monitoring method. The system includes:

[0213] The image processing module is based on the umbilical cord mesenchymal stem cell image, performs image size standardization, adjusts the aspect ratio of the image, executes noise removal, removes the background noise of the image, crops the cell area, removes the gray value area below the set threshold, and eliminates the noise points in the image through the gray value comparison operation, extracts the edge information of the cells, and generates a denoised cell image;

[0214] The feature extraction module is based on the denoised cell image, inputs the image into the residual network, processes the output features through convolution operations, adjusts the convolution kernel size to match the cell morphological changes, calculates the influence of the proliferation state on the cell morphology, extracts the features of the cell edge, and generates a cell edge feature map;

[0215] Based on the cell edge feature map, the boundary recognition module performs edge change and gray-scale comparison, identifies the boundary region between the cell and the background, extracts the edge position of the cell, executes region division and boundary extraction, combines the cell morphology information, conducts boundary value comparison analysis, and obtains the cell boundary interval value;

[0216] Based on the cell boundary interval value, the cycle analysis module classifies the cell cycle stages, analyzes the periodicity through the cell boundary change, determines whether the cell is in the division stage or the stationary stage, calculates the change in the cell morphology within the cycle, and obtains the cell cycle stage boundary map;

[0217] Based on the cell cycle stage boundary map, the deep learning optimization module conducts deep learning training, optimizes the hierarchical depth and learning rate of the neural network, adjusts the learning rate and weight of each layer, optimizes the image segmentation process, and generates the optimized cell segmentation result map.

[0218] The above is only the preferred embodiment of the present invention, and it is not intended to limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. Method for monitoring the preparation process of umbilical cord mesenchymal stem cells, characterized in that, It includes the following steps: S1: Based on the umbilical cord mesenchymal stem cell image, perform image size standardization, noise removal, cropping of the cell region, execute gray value comparison to remove the gray value region below the set threshold, apply a filtering method to eliminate background noise, remove noise points and extract cell edge information, and generate a denoised cell image; S2: Based on the denoised cell image, input the cell image into a residual network, process the output features of each layer through convolution operation weighted sum, apply a non-linear activation function to adjust the convolution kernel size to match the cell morphological changes, calculate the influence of the cell proliferation state on the morphological features, and generate a cell edge feature map; S3: Based on the cell edge feature map, perform edge change and gray value comparison, identify and extract the boundary of umbilical cord mesenchymal stem cells, execute region division and boundary extraction, combine cell morphological monitoring information, perform boundary value comparison analysis, and generate a cell boundary interval value; S4: Based on the cell boundary interval value, perform cell cycle stage classification, conduct periodic analysis through cell boundary changes, set a standard to divide the mitosis phase and the stationary phase, calculate the cell morphological changes within the cycle, and generate a cell cycle stage boundary map; S5: Based on the cell cycle stage boundary map, perform deep learning training, conduct cycle state subdivision and optimization, adjust the depth of the neural network layer and the learning rate, optimize the image segmentation process, adjust the learning rate and weight of each layer, and obtain an optimized segmentation result map.

2. The method for monitoring the preparation process of umbilical cord mesenchymal stem cells according to claim 1, characterized in that, The denoised cell image specifically includes the cell image, the noise removal region, and the cell edge information. The cell edge feature map includes convolution features, activation function adjustment, and morphological change influence. The cell boundary interval value includes edge change, gray value comparison, boundary extraction, and morphological monitoring. The cell cycle stage boundary map includes the mitosis phase, the stationary phase, and cell morphological changes. The optimized segmentation result map specifically includes cycle state optimization, network layer adjustment, and segmentation result.

3. The method for monitoring the preparation process of umbilical cord mesenchymal stem cells according to claim 2, wherein The specific steps for obtaining the denoised cell image are as follows: S101: According to the input umbilical cord mesenchymal stem cell image, adjust the size of all images by a unified scaling ratio. At the same time, based on the set threshold, compare the gray value of each pixel with the threshold for judgment, filter out the gray values less than the threshold, and generate a standardized image; S102: In the standardized image, use a high-pass or low-pass filter to filter the image, distinguish different noise frequencies, adjust the filtering intensity using the design parameters of the filter, and at the same time, through threshold determination, separate the cell region from other regions to establish a denoised cell region; S103: Based on the denoised cell region, locate the edge position by calculating the gray value difference between each pixel in the image and its neighboring pixels, compare the edge intensity using a preset threshold, and use the formula: Calculate the cell edge information and generate a denoised cell image; Among them, E represents the edge information amount, I i represents the gray value of the i-th pixel in the image, α is the edge sharpness coefficient, β is the edge smoothness adjustment coefficient, γ is the detail enhancement coefficient, and N is the total number of pixels in the image.

4. The method for monitoring the preparation process of umbilical cord mesenchymal stem cells according to claim 3, wherein, The specific steps for obtaining the cell edge feature map are as follows: S201: Based on the denoised cell image, input the cell image into the residual network, perform weighted processing on the feature maps of each layer through convolution operations, and at the same time filter the image multiple times through filters, weight the output features of the weighted differentiation layer, adjust the convolution kernel size of the differentiation layer, and calculate the convolution output of each layer to obtain the convolution result image; S202: Apply a non - linear activation function to adjust the convolution result image. According to the changes in cell morphology, adjust the non - linear effect of the output features based on the differential morphology of the cells. By adjusting the convolution kernel size, obtain a feature map that matches the cell morphology; S203: Based on the feature map that matches the cell morphology, according to the relationship between the proliferation state and cell morphology, analyze the changes in cell morphological features under different proliferation states. By calculating the proliferation rate and the amplitude of morphological changes, and combining the influence of cell proliferation on image features, adjust the feature map to reflect the changes in cell proliferation state, using the formula: Calculate the influence of cell proliferation on morphological features and generate a cell edge feature map; Among them, C represents the change value of the morphological feature map, and F i represents the gray value of the i-th pixel in the image, α is the coefficient of morphological change, β is the reference coefficient of morphological change, γ is the influence coefficient of the proliferation rate, δ is the adjustment factor, and M is the total number of pixels in the image.

5. The method for monitoring the preparation process of umbilical cord mesenchymal stem cells according to claim 4, wherein The specific steps for obtaining the cell boundary interval value are as follows: S301: Based on the cell edge feature map, extract the gray - scale gradient values of the cell edge changes. By comparing the gray - scale gradient values with the target threshold, screen the pixel points that conform to the cell boundary features to generate a cell edge pixel feature set; S302: Call the cell edge pixel feature set, calculate the continuity value of the edge change according to the directionality of the edge change and the gray - scale distribution uniformity index, perform regional division, and generate a regional boundary feature set; S303: Call the regional boundary feature set, perform difference value operations between the boundary feature parameters, using the formula: Calculate the change intensity of the regional boundary, compare and extract the maximum change intensity value to generate the cell boundary interval value; Among them, B i represents the boundary interval value, G j represents the gray value of the j-th boundary pixel point, G j+1 represents the gray value of the adjacent pixel point, D j represents the distance between pixel points, is the average value of all distances, α is the adjustment coefficient of gray change, β is the smoothing parameter of the set of regional boundary characteristics, and n is the number of boundary pixel points.

6. The method for monitoring the preparation process of umbilical cord mesenchymal stem cells according to claim 5, characterized in that, The specific steps for obtaining the cell cycle stage boundary map are as follows: S401: Based on the cell boundary interval value, through cell cycle stage classification, perform periodic analysis on the cell boundary changes, calculate the difference value of the boundary changes within the cell cycle, set the standard division between the mitosis stage and the stationary stage, and generate the cell cycle stage boundary map; S402: Call the cell cycle stage boundary map, combine the time characteristics of the boundary changes, calculate the cell morphological changes in different differential stages, analyze the change amplitude in multiple stages within the cycle, and generate the cell boundary distribution map of different differential stages within the cycle; S403: In the cell boundary distribution map of different differential stages within the cycle, use the formula: Calculate the boundary change intensity of multiple cell cycle stages, optimize the division of multiple stages within the cycle, and generate the cell cycle stage boundary map; Among them, C s represents the change intensity of the cycle stage boundary, P k represents the boundary value of the k-th cell stage, Y k represents the boundary position of the k-th stage, Y k+1 represents the boundary position of adjacent stages, is the mean value of the boundary position, β is the adjustment coefficient, γ t is the time adjustment coefficient, and m is the total number of stages in the cell cycle.

7. The method for monitoring the preparation process of umbilical cord mesenchymal stem cells according to claim 6, characterized in that, The specific steps for obtaining the optimized segmentation result map are as follows: S501: Based on the cell cycle stage boundary map, perform cycle state subdivision and optimization through deep - learning training. Combine the image data and the neural network model, adjust the depth of the neural network layer and the learning rate to generate an optimized neural network model; S502: Invoke the optimized neural network model, perform backpropagation and optimize the image segmentation process by adjusting the learning rate and weights of each layer, calculate the gradient value of the weights of each layer, and generate an optimized segmentation strategy; S503: In the application of the optimized segmentation strategy, use the formula: Calculate the error function of each neural network layer, and perform parameter adjustment to generate an optimized segmentation result map; Among them, L represents the loss function value, Z i represents the true value of the i-th image pixel, represents the predicted value, λ is the regularization coefficient, α j is the learning rate adjustment coefficient for layer j, W j represents the weight value of the j-th layer, n is the total number of pixel points, and m is the number of network layers.

8. Umbilical cord mesenchymal stem cell preparation process monitoring system, characterized in that, According to the method for monitoring the preparation process of umbilical cord mesenchymal stem cells according to any one of claims 1-7, the system includes: The image processing module performs image size standardization based on the umbilical cord mesenchymal stem cell image, adjusts the aspect ratio of the image, performs noise removal, removes the background noise of the image, crops the cell region, removes the gray value region below the set threshold, eliminates the noise points in the image through gray value comparison operations, extracts the edge information of the cells, and generates a denoised cell image; The feature extraction module inputs the image to the residual network based on the denoised cell image, processes the output features through convolution operations, adjusts the convolution kernel size to match the cell morphology changes, calculates the influence of the proliferation state on the cell morphology, extracts the features of the cell edges, and generates a cell edge feature map; The boundary recognition module performs edge change and gray value comparison based on the cell edge feature map, recognizes the boundary region between the cells and the background, extracts the edge positions of the cells, performs region division and boundary extraction, combines the cell morphology information, and performs boundary value comparison analysis to obtain the cell boundary interval value; The cycle analysis module classifies the cell cycle stages based on the cell boundary interval value, analyzes the periodicity through the cell boundary changes, determines whether the cells are in the division phase or the stationary phase, calculates the changes in the cell morphology within the cycle, and obtains a cell cycle stage boundary map; The deep learning optimization module performs deep learning training based on the cell cycle stage boundary map, optimizes the layer depth and learning rate of the neural network, adjusts the learning rate and weights of each layer, optimizes the image segmentation process, and generates an optimized cell segmentation result map.

Images