Stem cell quality evaluation system and method

Abstract

The invention relates to the technical field of stem cell biology, and discloses a stem cell quality evaluation system and method, and the system comprises a data collection module which collects local environment data and image data of stem cells in a preset range; the morphological recognition module is used for obtaining cellular morphological data of all stem cells in a preset range through a cellular morphological recognition model; the graph structure data construction module is used for constructing graph structure data; the score marking module is used for measuring the multiplication capacity, the genetic stability capacity and the surface marker expression capacity of the stem cells after the preset duration, and obtaining the quality score of the stem cells through manual marking; the model training module is used for training a stem cell quality scoring model through an optimization algorithm; according to the method, the cellular morphology data of the stem cells are extracted through the multi-branch network of the cellular morphology recognition model, and the interaction among the stem cells is comprehensively considered through the stem cell quality scoring model, so that the accuracy of stem cell quality evaluation is improved.

Description

Technical Field

[0001] The present invention relates to the field of stem cell biotechnology, and more specifically, it relates to a stem cell quality evaluation system and method. Background Art

[0002] Stem cells are a special type of cells with self-renewal ability and multi-directional differentiation potential, and have broad application prospects in the fields of regenerative medicine, tissue engineering, and cell therapy. In order to ensure their safety and effectiveness in the clinical and industrialization processes, stem cell quality evaluation has become a key link. Traditional stem cell quality evaluation mainly relies on a series of experimental detection methods, including but not limited to determination of cell proliferation ability (such as population doubling time), detection of genetic stability (such as chromosome karyotype analysis), and detection of surface marker expression (such as flow cytometry detection of CD molecules), etc. Although the above methods can reflect the functional state of stem cells, they have limitations such as long detection period, high experimental cost, and high technical threshold, and it is difficult to achieve real-time monitoring of the state of stem cells.

[0003] With the development of artificial intelligence, there are existing methods for quality assessment of stem cells through intelligent image analysis means. For example, cell image data is collected through high-resolution microscopy imaging, and image features of the image data are extracted by combining technical means such as multi-scale convolution and attention mechanism, and then converted into stem cell quality scores through non-linear mapping methods (such as multi-layer perceptron, support vector machine, etc.). However, the above methods generally analyze individual stem cells as isolated individuals, ignoring the interactions between stem cells, such as paracrine effects (stem cells secrete cytokines to affect the differentiation direction and proliferation rate of neighboring cells), cell contact regulation (cell contact also affects the proliferation rate of neighboring cells), etc., resulting in deviations in the quality assessment of stem cells by the above methods. Summary of the Invention

[0004] The present invention provides a stem cell quality evaluation system and method to solve the technical problems in the above background art.

[0005] The present invention provides a stem cell quality evaluation system, including: A data acquisition module, which is used to acquire local environment data and image data of stem cells within a preset range; The local environment data includes: pH value, temperature, nutrient concentration, and the number of stem cells within a preset range; A morphology recognition module, which is used to obtain cell morphology data of all stem cells within a preset range through a cell morphology recognition model; The cell morphology data includes: cell area, aspect ratio, vacuole rate, edge sharpness, and whether adherent; A graph structure data construction module, which is used to construct graph structure data according to the cell morphology data of all stem cells within a preset range; The graph structure data consists of nodes and edges between the nodes; The nodes include core nodes and neighbor nodes. The core nodes are represented by the cell morphology data of the stem cells at the center of the preset range, and the neighbor nodes are represented by the cell morphology data of other stem cells; Constructing the edges between the nodes must meet the preset conditions; A scoring and annotation module, which is used to measure the proliferation ability, genetic stability ability, and surface marker expression ability of stem cells after a preset time period, and obtain the quality score of the stem cells through manual annotation; A model training module, which is used to use the graph structure data of the stem cells and the local environment data within the preset range as sample data, and use the quality score of the stem cells after the preset time period as a sample label, and train the stem cell quality scoring model using the sample data and the sample label through an optimization algorithm.

[0006] Further, the preset time period is a custom parameter, and the preset range represents a circular area generated with the stem cell as the center and a preset distance as the radius, where the preset distance is a custom parameter.

[0007] Further, the preset conditions include: the Euclidean distance between the corresponding stem cells is less than the distance threshold, where the distance threshold is a custom parameter; the correlation coefficient between the corresponding stem cells is greater than or equal to the coefficient threshold, where the coefficient threshold is a custom parameter.

[0008] Further, the cell morphology recognition model consists of 5 branches with the same structure but not sharing weight parameters. The inputs of the 5 branches are all the image data of the stem cells, and the outputs are the cell area, aspect ratio, vacuolation rate, edge sharpness, and whether they adhere to the wall respectively; Each branch consists of a region division layer, a linear projection layer, a position encoding layer, a feature extraction layer, and a first classifier; The region division layer is used to divide the image data into N square regions, and each square region is represented by a first vector with a dimension number of P²C, where N = H×W / P², H, W, and C respectively represent the height, width, and number of channels of the image data, P represents the side length of the square region, and P is a custom parameter; The linear projection layer is used to convert each square region into a second vector representation with a dimension number of D, where D is a custom parameter. The calculation formula of the linear projection layer is as follows: ; Where and respectively represent the second vector and the first vector of the i-th square region, represents the weight matrix, represents the bias vector, and e represents the natural constant; The position encoding layer is used to obtain a third vector by embedding the position vector of the second vector for each square region. The calculation formula of the position vector is as follows: ; where L represents the maximum value of the self-increment order and is a custom parameter, represents the two-dimensional values of the k-th self-increment order of the position vector of the i-th square region, p represents the normalized position number, and norm represents the Min-Max normalization method; The feature extraction layer is used to extract features from the sequence composed of the third vectors of N square regions to obtain a fourth vector, and use the fourth vector as the input of the first classifier.

[0009] Furthermore, the calculation formula of the feature extraction layer includes: ; ; ; where represents the fourth vector output by the feature extraction layer, M represents the attention matrix, and represent the weight matrix and bias vector corresponding to the attention matrix respectively. Q, K, and V represent the query matrix, key matrix, and value matrix respectively, and are all obtained by multiplying the sequence composed of the third vectors of N square regions by the corresponding weight matrix, and their sizes are all N×A, where A is a custom parameter, represents the masking operation on the query matrix and the key matrix to obtain the masking matrix, , and represent the elements of the n-th row and a-th column of the masking matrix, query matrix, and key matrix respectively, hash represents the hash function, T represents the transpose operation, represents element-wise multiplication, Swish represents the Swish activation function, and softmax represents the softmax activation function.

[0010] Furthermore, the number of newly proliferated cells is obtained by the cell counting method as the proliferation ability, the number of newly proliferated cells with normal chromosomes is obtained by karyotype analysis as the genetic stability ability, and the number of newly proliferated cells with positive flow cytometry results is obtained by flow cytometry as the surface marker expression ability.

[0011] Furthermore, the stem cell quality scoring model consists of a graph structure data analysis layer, a splicing layer, and a second classifier; The graph structure data analysis layer is used to update the graph structure data of stem cells; The splicing layer is used to splice and normalize the vector of the updated core node with the local environment data within a preset range to obtain a combined vector, and the number of dimensions of the combined vector is a custom parameter; The second classifier inputs the combined vector, and the class space of the second classifier represents the quality score of the stem cells; The graph structure data analysis layer is constructed based on the graph attention network model.

[0012] Furthermore, training the stem cell quality score model through an optimization algorithm includes the following steps: Step S201, randomly generate the parameters of the stem cell quality score model as the encoding of the individuals in the initial population; Step S202, calculate the loss values of all individuals in the initial population through the loss function; The calculation formula of the loss function is the mean square error between the value output by the stem cell quality score model using the parameters corresponding to the encoding of the individual and the sample label; Step S203, generate a non-linear adjustment factor according to the current iteration number and the maximum iteration number, and judge whether the non-linear adjustment factor is less than the preset factor threshold. If so, enter Step S204; otherwise, enter Step S205; The calculation formula of the non-linear adjustment factor Factor is as follows: ; where t represents that the current iteration number is t, represents the maximum iteration number, and the preset factor threshold is a custom parameter; Step S204, increment the current iteration number by 1, and update the encoding of all individuals in the initial population through the first update strategy; Step S205, increment the current iteration number by 1, and update the encoding of all individuals in the initial population through the second update strategy; Step S206, judge whether the current iteration number is greater than or equal to the maximum iteration number or the minimum value of the loss values of the individuals in the initial population is less than or equal to the preset loss threshold. If so, terminate the iteration, and use the encoding of the individual with the smallest loss value as the parameters of the stem cell quality score model; otherwise, return to Step S202 to continue execution; where both the maximum iteration number and the preset loss threshold are custom parameters.

[0013] Furthermore, the calculation formula of the optimization algorithm includes: The calculation formula of the first update strategy includes: ; ; where 1 ≤ g ≤ G, 1 ≤ t ≤ , and respectively represent the encodings of the g-th individual at the current iteration t and t + 1, represents the encoding of a randomly selected individual at the current iteration t, and R represents the correlation coefficient, represents the first random number with a value range between 0 and 1; The calculation formula of the second update strategy is as follows: ; where represents the encoding of the individual with the smallest loss value in the initial population at the current iteration t, represents the second random number with a value range between 0 and 1.

[0014] The present invention provides a method for evaluating the quality of stem cells, comprising the following steps: Step S301, collecting local environmental data and image data of stem cells within a preset range; Step S302, obtaining cell morphology data of all stem cells within a preset range through a cell morphology recognition model; Step S303, constructing graph structure data based on the cell morphology data of all stem cells within a preset range; Step S304, measuring the proliferation ability, genetic stability ability, and surface marker expression ability of stem cells after a preset time period, and obtaining a quality score of the stem cells through manual annotation; Step S305, using the graph structure data of the stem cells and the local environmental data within a preset range as sample data, and using the quality score of the stem cells after a preset time period as a sample label, and training a stem cell quality scoring model through an optimization algorithm using the sample data and the sample label.

[0015] The beneficial effects of the present invention are as follows: The present invention extracts the cell morphology data of stem cells through the multi-branch network of the cell morphology recognition model, and comprehensively considers the interaction between stem cells through the stem cell quality scoring model, thereby improving the accuracy of stem cell quality assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a schematic diagram of a stem cell quality evaluation system of the present invention; Figure 2 is a flowchart of training a stem cell quality scoring model through an optimization algorithm of the present invention; Figure 3 is a flowchart of a method for evaluating the quality of stem cells of the present invention.

[0017] In the figure: data acquisition module 101, morphology recognition module 102, graph structure data construction module 103, scoring annotation module 104, model training module 105. Detailed implementation manners

[0018] Now, the subject matter described herein will be discussed with reference to exemplary implementation manners. It should be understood that discussing these implementation manners is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described relative to some examples can also be combined in other examples.

[0019] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by those of ordinary skill in the art to which the present invention pertains. The "first", "second", and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are only used to distinguish different components. Words such as "including" or "comprising" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. Words such as "connected" or "coupled" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Upper", "lower", "left", "right", etc. are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0020] As Figures 1 to 3 shown, a stem cell quality evaluation system includes: A data acquisition module 101, which is used to acquire local environment data and image data of stem cells within a preset range; The local environment data includes: pH value, temperature, nutrient concentration, and the number of stem cells within a preset range; A morphology recognition module 102, which is used to obtain cell morphology data of all stem cells within a preset range through a cell morphology recognition model; The cell morphology data includes: cell area, aspect ratio, vacuolation rate, edge sharpness, and whether it adheres to the wall; A graph structure data construction module 103, which is used to construct graph structure data according to the cell morphology data of all stem cells within a preset range; The graph structure data consists of nodes and edges between the nodes; The nodes include core nodes and neighbor nodes. The core nodes are represented by the cell morphological data of the stem cells at the center of the preset range, while the neighbor nodes are represented by the cell morphological data of other stem cells; The edges constructed between the nodes must meet the preset conditions; The scoring and annotation module 104 is used to measure the proliferation ability, genetic stability ability, and surface marker expression ability of the stem cells after a preset time period, and obtain the quality score of the stem cells through manual annotation; The model training module 105 is used to use the graph structure data of the stem cells and the local environment data within the preset range as sample data, and the quality score of the stem cells after a preset time period as a sample label, and train the stem cell quality scoring model using the sample data and the sample label through an optimization algorithm.

[0021] In an embodiment of the present invention, the preset time period is a custom parameter. For example, the preset time period is set to 48 hours. The preset range represents a circular area generated with the stem cell as the center and a preset distance as the radius, where the preset distance is a custom parameter. For example, the preset distance is set to 200 micrometers.

[0022] In an embodiment of the present invention, the preset conditions include: the Euclidean distance between the corresponding stem cells is less than the distance threshold, where the distance threshold is a custom parameter. For example, the distance threshold is set to 50 micrometers; the correlation coefficient between the corresponding stem cells is greater than or equal to the coefficient threshold, where the coefficient threshold is a custom parameter. For example, the coefficient threshold is set to 0.7, and the correlation coefficient can be any one of the Pearson correlation coefficient, the Spearman correlation coefficient, or the cosine similarity.

[0023] In an embodiment of the present invention, the cell morphology recognition model is composed of 5 branches with the same structure but without sharing weight parameters. The inputs of the 5 branches are all the image data of the stem cells, and the outputs are the cell area, aspect ratio, vacuole rate, edge sharpness, and whether they adhere to the wall respectively; Each branch is composed of a region division layer, a linear projection layer, a position encoding layer, a feature extraction layer, and a first classifier; The region division layer is used to divide the image data into N square regions, and each square region is represented by a first vector with a dimension number of P²C, where N = H×W / P², and H, W, and C respectively represent the height, width, and number of channels of the image data, and P represents the side length of the square region, and P is a custom parameter. For example, P is set to H / 16; It should be noted that before inputting into the cell morphology recognition model, the image data of the stem cells can also be preprocessed, such as denoising, unifying the size, etc.; The linear projection layer is used to convert each square region into a second vector representation with a dimensionality of D, where D is a custom parameter. For example, D is set to 16. The calculation formula of the linear projection layer is as follows: ; where and represent the second vector and the first vector of the i-th square region respectively, represents the weight matrix, has a size of P²C×D, represents the bias vector, has a dimensionality of D, and e represents the natural constant; The position encoding layer is used to embed the position vector into the second vector of each square region to obtain a third vector. The calculation formula of the position vector is as follows: ; where L represents the maximum value of the self-increment order and is a custom parameter. For example, L is set to 8, represents the two-dimensional values of the k-th self-increment order of the position vector of the i-th square region, p represents the normalized position number, norm represents the Min-Max normalization method, the dimensionality of the position vector is 2L, and the dimensionality of the third vector is D + 2L; It should be noted that the position vector can also be obtained through fixed sine-cosine position encoding. The specific calculation formulas include: ; ; where 1 ≤ i ≤ N, 1 ≤ j ≤ D, and represent the 2j-th and (2j + 1)-th dimensional values of the position vector of the i-th square region respectively, D represents the dimensionality of the second vector, that is, the dimensionality of the position vector is the same as that of the second vector; The feature extraction layer is used to extract features from the sequence composed of the third vectors of N square regions to obtain a fourth vector, and use the fourth vector as the input of the first classifier.

[0024] It should be noted that the first classifiers of the above 5 branches are all constructed based on multi-layer perceptrons. The activation functions of the first classifiers with the category spaces of cell area, aspect ratio, vacuole ratio, and edge sharpness are all PReLU activation functions, and the activation function of the first classifier with the category space of whether adherent is the Sigmoid activation function, which will not be elaborated here.

[0025] It should be noted that the cell area is related to the cell's proliferation ability; the aspect ratio reflects the morphological symmetry of the cell. Normal stem cells usually maintain a symmetric morphology, and an irregular aspect ratio may indicate the differentiation or stress state of the stem cells; the vacuole rate represents the proportion of the transparent area of the cell. A higher vacuole rate may indicate that the cell is under a certain physiological stress, such as nutrient deficiency, oxidative stress, etc., which has a certain impact on the health and differentiation ability of the cell; the edge sharpness reflects the integrity and structural health of the cell membrane; the adhesion ability of stem cells reflects their vitality and proliferation state. Usually, healthy stem cells can adhere to the culture medium bottom.

[0026] In one embodiment of the present invention, the calculation formula of the feature extraction layer includes: ; where represents the fourth vector output by the feature extraction layer, M represents the attention matrix, with a size of N×A, and respectively represent the weight matrix and bias vector corresponding to the attention matrix, has a size of 1×N, has a dimension number of A, then the dimension number of the fourth vector is A, and Swish represents the Swish activation function; ; ; where Q, K, and V respectively represent the query matrix, key matrix, and value matrix, and are all obtained by multiplying the sequence (N×D) of the third vectors of N square regions by the corresponding weight matrix, and their sizes are all N×A, that is, the size of the weight matrix is D×A, where A is a custom parameter, for example, A is set to 16, represents the masking operation on the query matrix and the key matrix to obtain the masking matrix, that is, the size of the masking matrix is N×A, , and respectively represent the element values of the nth row and the ath column of the masking matrix, query matrix, and key matrix, hash represents the hash function, T represents the transpose operation, represents element-wise multiplication, and softmax represents the softmax activation function.

[0027] It should be noted that the parameters (weight matrix, bias vector) in the cell morphology recognition model are all learnable hyperparameters, and the sample labels used to train the cell morphology recognition model are all obtained through manual annotation. During the training process, the difference between the sample label and the value output by the cell morphology recognition model can be specified as the loss function, and the parameters in the cell morphology recognition model can be updated backward using a gradient optimizer (such as AdaGrad, RMSProp, etc.) to minimize the loss and complete the convergence of the model. In addition, the addition of the masking operation is mainly for information screening to reduce the data calculation amount.

[0028] In one embodiment of the present invention, the number of newly proliferated cells obtained by cell counting is used as the proliferation ability, the number of newly proliferated cells with normal chromosomes obtained by karyotype analysis is used as the genetic stability ability, and the number of newly proliferated cells with a positive flow cytometry result obtained by flow cytometry is used as the surface marker expression ability.

[0029] It should be noted that in addition to manual annotation, automatic annotation can also be directly achieved through normalization and weighting operations, and the corresponding weighting coefficients can be set artificially. In addition, the flow cytometry result obtained by flow cytometry is related to the type of CD molecule. Positive indicates that the stem cell expresses the marker (CD molecule), and negative indicates that the stem cell does not express the marker. For example, the markers that mesenchymal stem cells should be positive for include CD73, CD90, etc., while CD34, CD45, etc. should be negative. If CD90 is positive, it indicates that the quality of the stem cell is good, and if CD45 is positive, it indicates that the quality of the stem cell is abnormal, which will not be elaborated here.

[0030] In one embodiment of the present invention, the stem cell quality scoring model consists of a graph structure data analysis layer, a splicing layer, and a second classifier; The graph structure data analysis layer is used to update the graph structure data of the stem cells; The splicing layer is used to splice and normalize the vector of the updated core node with the local environment data within a preset range to obtain a combined vector, and the number of dimensions of the combined vector is a custom parameter. For example, the number of dimensions of the combined vector is set to 32; The second classifier inputs the combined vector, and the category space of the second classifier represents the quality score of the stem cell; It should be noted that the activation function of the second classifier can also be designed as a PReLU activation function, and the training methods of the stem cell quality scoring model and the cell morphology recognition model are the same.

[0031] In one embodiment of the present invention, the graph structure data analysis layer is constructed based on the GAT (Graph Attention Network) model, or can also be constructed based on the GCN (Graph Convolutional Network) model, which will not be elaborated here.

[0032] In one embodiment of the present invention, as Figure 2 shown, training the stem cell quality scoring model through an optimization algorithm includes the following steps: Step S201, randomly generate the parameters of the stem cell quality scoring model as the encoding of the individuals in the initial population; Step S202, calculate the loss values of all individuals in the initial population through the loss function; The calculation formula of the loss function is the mean square error between the value output by the stem cell quality scoring model using the parameters corresponding to the encoding of the individual and the sample label; Step S203, generate a non-linear adjustment factor according to the current iteration number and the maximum iteration number, and determine whether the non-linear adjustment factor is less than the preset factor threshold. If so, enter step S204; otherwise, enter step S205; The calculation formula of the non-linear adjustment factor Factor is as follows: ; where t represents that the current iteration number is t, represents the maximum iteration number, the preset factor threshold is a user-defined parameter. For example, the preset factor threshold is set to 1; Step S204, increment the current iteration number by 1, and update the encoding of all individuals in the initial population through the first update strategy; Step S205, increment the current iteration number by 1, and update the encoding of all individuals in the initial population through the second update strategy; Step S206, determine whether the current iteration number is greater than or equal to the maximum iteration number or the minimum value of the loss values of the individuals in the initial population is less than or equal to the preset loss threshold. If so, terminate the iteration, and use the encoding of the individual with the minimum loss value as the parameters of the stem cell quality scoring model; otherwise, return to step S202 to continue execution; where both the maximum iteration number and the preset loss threshold are user-defined parameters. For example, the maximum iteration number is set to 100, and the preset loss threshold is set to 0.01.

[0033] In one embodiment of the present invention, the calculation formula of the optimization algorithm includes: The calculation formula of the first update strategy includes: ; ; where 1 ≤ g ≤ G, 1 ≤ t ≤ , and respectively represent the encoding of the g-th individual at the current iteration number t and t + 1, represents the encoding of a random individual at the current iteration number t, R represents the correlation coefficient, Represents a first random number with a value range between 0 and 1; The calculation formula of the second update strategy is as follows: ; Where represents the encoding of the individual with the smallest loss value in the initial population when the current iteration number is t, represents a second random number with a value range between 0 and 1.

[0034] In an embodiment of the present invention, as Figure 3 shown, a method for evaluating the quality of stem cells includes the following steps: Step S301, collect local environment data and image data of stem cells within a preset range; Step S302, obtain cell morphology data of all stem cells within a preset range through a cell morphology recognition model; Step S303, construct graph structure data based on the cell morphology data of all stem cells within a preset range; Step S304, measure the proliferation ability, genetic stability ability, and surface marker expression ability of stem cells after a preset time duration, and obtain the quality score of stem cells through manual annotation; Step S305, use the graph structure data of stem cells and the local environment data within a preset range as sample data, and use the quality score of stem cells after a preset time duration as a sample label, and train a stem cell quality score model using the sample data and the sample label through an optimization algorithm.

[0035] The above has described the embodiments of this embodiment, but this embodiment is not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of this embodiment.

Claims

1. A stem cell quality evaluation system, characterized in that, Including: A data acquisition module, which is used to acquire the local environment data and image data of stem cells within a preset range; The local environment data includes: pH value, temperature, nutrient concentration, and the number of stem cells within a preset range; A morphology recognition module, which is used to obtain the cell morphology data of all stem cells within a preset range through a cell morphology recognition model; The cell morphology data includes: cell area, aspect ratio, vacuolation rate, edge sharpness, and whether it adheres to the wall; A graph structure data construction module, which is used to construct graph structure data according to the cell morphology data of all stem cells within a preset range; The graph structure data consists of nodes and edges between the nodes; The nodes include core nodes and neighbor nodes. The core nodes are represented by the cell morphology data of the stem cells at the center of the preset range, and the neighbor nodes are represented by the cell morphology data of other stem cells; Constructing the edges between the nodes must meet preset conditions; A scoring and annotation module, which is used to measure the proliferation ability, genetic stability ability, and surface marker expression ability of stem cells after a preset time period, and obtain the quality score of stem cells through manual annotation; A model training module, which is used to use the graph structure data of stem cells and the local environment data within a preset range as sample data, and use the quality score of stem cells after a preset time period as a sample label, and train a stem cell quality score model through an optimization algorithm using the sample data and the sample label.

2. The stem cell quality evaluation system according to claim 1, wherein The preset time period is a custom parameter, and the preset range represents a circular area generated with a stem cell as the center and a preset distance as the radius, where the preset distance is a custom parameter.

3. The stem cell quality evaluation system according to claim 1, characterized in that The preset conditions include: the Euclidean distance between corresponding stem cells is less than a distance threshold, where the distance threshold is a custom parameter; the correlation coefficient between corresponding stem cells is greater than or equal to a coefficient threshold, where the coefficient threshold is a custom parameter.

4. A stem cell quality evaluation system according to claim 1, wherein The cell morphology recognition model consists of 5 branches with the same structure but without sharing weight parameters. The inputs of the 5 branches are all the image data of stem cells, and the outputs are the cell area, aspect ratio, vacuolation rate, edge sharpness, and whether it adheres to the wall respectively; Each branch consists of a region division layer, a linear projection layer, a position encoding layer, a feature extraction layer, and a first classifier; The region division layer is used to divide the image data into N square regions, and each square region is represented by a first vector with a dimension number of P²C, where N = H×W / P², and H, W, and C respectively represent the height, width, and number of channels of the image data, and P represents the side length of the square region, and P is a custom parameter; The linear projection layer is used to convert each square region into a second vector with a dimension number of D, where D is a custom parameter. The calculation formula of the linear projection layer is as follows: ； where and respectively represent the second vector and the first vector of the i-th square region, represents the weight matrix, represents the bias vector, and e represents the natural constant; The position encoding layer is used to embed a position vector into the second vector of each square region to obtain a third vector. The calculation formula of the position vector is as follows: ； where L represents the maximum value of the increment order and is a user-defined parameter, denotes the two-dimensional values of the k-th increment order of the position vector of the i-th square region, p represents the normalized position number, and norm represents the Min-Max normalization method; The feature extraction layer is used to perform feature extraction on the sequence composed of the third vectors of N square regions to obtain a fourth vector, and use the fourth vector as the input of the first classifier.

5. The stem cell quality evaluation system according to claim 4, wherein The calculation formula of the feature extraction layer includes: ； ； ； Among them represents the fourth vector output by the feature extraction layer, M represents the attention matrix, and respectively represent the weight matrix and bias vector corresponding to the attention matrix. Q, K, and V respectively represent the query matrix, key matrix, and value matrix, which are all obtained by multiplying the sequence of the third vectors of N square regions by the corresponding weight matrices, and their sizes are all N×A, where A is a custom parameter. represents obtaining a masked matrix by performing a masking operation on the query matrix and the key matrix. 、 and respectively represent the element values of the nth row and a-th column of the masked matrix, query matrix, and key matrix. hash represents the hash function, and T represents the transpose operation. represents element-wise multiplication. Swish represents the Swish activation function, and softmax represents the softmax activation function.

6. The stem cell quality evaluation system according to claim 1, characterized in that, The number of newly proliferated cells obtained by cell counting method is used as the proliferation ability, the number of newly proliferated cells with normal chromosomes obtained by karyotype analysis is used as the genetic stability ability, and the number of newly proliferated cells with positive flow cytometry results obtained by flow cytometry is used as the surface marker expression ability.

7. The stem cell quality evaluation system according to claim 1, wherein The stem cell quality scoring model consists of a graph structure data analysis layer, a splicing layer, and a second classifier; The graph structure data analysis layer is used to update the graph structure data of stem cells; The splicing layer is used to splice and normalize the vector of the updated core node with the local environment data within a preset range to obtain a combined vector, and the number of dimensions of the combined vector is a custom parameter; The second classifier inputs the combined vector, and the category space of the second classifier represents the quality score of stem cells; The graph structure data analysis layer is constructed based on the graph attention network model.

8. The stem cell quality evaluation system according to claim 1, characterized in that The stem cell quality scoring model is trained by an optimization algorithm, including the following steps: Step S201, randomly generate the parameters of the stem cell quality scoring model as the encoding of the individuals in the initial population; Step S202, calculate the loss values of all individuals in the initial population through the loss function; The calculation formula of the loss function is the mean square error between the value output by the stem cell quality scoring model using the parameters corresponding to the encoding of the individual and the sample label; Step S203, generate a non-linear adjustment factor according to the current iteration number and the maximum iteration number, and judge whether the non-linear adjustment factor is less than the preset factor threshold. If so, go to step S204, otherwise go to step S205; The calculation formula of the non-linear adjustment factor Factor is as follows: ； where t represents that the current iteration number is t, represents the maximum iteration number, and the preset factor threshold is a custom parameter; Step S204, increment the current iteration number by 1, and update the encoding of all individuals in the initial population through the first update strategy; Step S205, increment the current iteration number by 1, and update the encoding of all individuals in the initial population through the second update strategy; Step S206, judge whether the current iteration number is greater than or equal to the maximum iteration number or the minimum value of the loss values of the individuals in the initial population is less than or equal to the preset loss threshold. If so, terminate the iteration, and use the encoding of the individual with the minimum loss value as the parameters of the stem cell quality scoring model. Otherwise, return to step S202 to continue execution; Wherein both the maximum iteration number and the preset loss threshold are custom parameters.

9. The stem cell quality evaluation system according to claim 8, wherein The calculation formula of the optimization algorithm includes: The calculation formula of the first update strategy includes: ； ； where 1 ≤ g ≤ G, 1 ≤ t ≤ , and represent the encodings of the g-th individual at the current iteration t and t+1 respectively, represents the encoding of a random individual at the current iteration t, R represents the correlation coefficient, represents the first random number with a value range between 0 and 1; The calculation formula of the second update strategy is as follows: ； Among them represents the encoding of the individual with the smallest loss value in the initial population when the current iteration number is t, represents the second random number with a value range between 0 and 1.

10. A method for evaluating the quality of stem cells, characterized in that, Execute a stem cell quality evaluation system according to any one of claims 1 to 9, including the following steps: Step S301, collect the local environment data and image data of stem cells within a preset range; Step S302, obtain the cell morphology data of all stem cells within a preset range through the cell morphology recognition model; Step S303, construct graph structure data according to the cell morphology data of all stem cells within a preset range; Step S304, measure the proliferation ability, genetic stability ability, and surface marker expression ability of stem cells after a preset time period, and obtain the quality score of stem cells through manual annotation; Step S305: Use the graph structure data of the stem cells and the local environment data within a preset range as sample data, and use the quality score of the stem cells after a preset duration as a sample label. Train a stem cell quality score model using the sample data and the sample label through an optimization algorithm.

Images