Method and apparatus for evaluating confluency of stem cell expansion culture

Abstract

The application relates to the technical field of stem cell culture, and provides a fusion degree evaluation method and equipment for stem cell expansion culture, which comprises the following steps: calculating a microenvironment feature by using a microenvironment parameter and a stem cell fusion degree parameter; obtaining a time perception label by performing time perception based on a two-dimensional image, and obtaining a fusion degree prediction label by performing fusion degree classification; fusing a two-dimensional image and a three-dimensional image based on the time perception label and the fusion degree prediction label to obtain an enhanced stem cell image; performing time perception on the enhanced stem cell image to obtain an enhanced time perception label, and performing fusion degree evaluation to obtain an enhanced fusion degree prediction label; and performing fusion degree evaluation by using a fusion degree evaluation model based on the microenvironment feature, the time perception label, the fusion degree prediction label, the enhanced time perception label and the enhanced fusion degree prediction label, so as to obtain a fusion degree evaluation value. The method can improve the accuracy of fusion degree evaluation.

Description

A method and equipment for assessing the fusion degree of stem cell expansion culture Technical Field

[0001] This application relates to the field of stem cell culture technology, and in particular to a method and device for evaluating the fusion degree of stem cell expansion culture. Background Technology

[0002] In today's biomedical field, cell therapy, especially stem cell therapy, has shown enormous potential, and its vigorous development has become one of the key industries of concern to various countries. Stem cells possess unique self-renewal and multi-directional differentiation capabilities, which makes them promising for applications in many important areas such as tissue engineering, regenerative medicine, and drug screening.

[0003] Stem cell confluence (the ratio of cell coverage area to the surface area of ​​the culture dish) directly affects stem cell metabolic activity. Assessing stem cell confluence in expanded cultures helps maintain healthy stem cell expansion and prevents differentiation and cell death. At low confluence, cells have lower metabolic levels, consume fewer nutrients in the culture medium, and do not require frequent medium changes. At moderate confluence, metabolic levels increase, requiring more nutrients, and metabolic waste accumulates more rapidly. At high confluence, cell density is high, nutrient demand increases, and metabolic waste accumulates even faster, necessitating more frequent medium changes. Effectively assessing stem cell confluence in expanded cultures remains a significant challenge.

[0004] Besides relying on microscopic observation and manual analysis, image recognition technology plays a crucial role in current methods. These machine learning-based image recognition technologies overcome the drawbacks of traditional manual methods, such as strong subjectivity, significant time consumption, and difficulty in handling large-scale data processing. However, some shortcomings remain: First, current methods do not consider microenvironmental changes caused by medium changes. These changes typically affect the short-term growth trend of cell fusion, which is of significant reference value for cell fusion assessment. Second, current methods generally use whole two-dimensional (2D) stem cell images for analysis, failing to consider the topological structure and block information aggregation between image blocks. This results in a lack of connection and constraints between image block features and information aggregation in time-aware tasks and cell fusion assessment tasks, leading to incomplete or biased assessment results. Third, current methods often perform image feature analysis based on 2D images, resulting in a single feature dimension and ignoring the interaction between 2D images and the limited and valuable three-dimensional (3D) images. Therefore, current stem cell fusion assessment methods suffer from low accuracy. Summary of the Invention

[0005] This application provides a method and device for assessing the fusion degree of stem cell expansion culture, which can solve the problem of low accuracy in fusion degree assessment.

[0006] In a first aspect, this application provides a method for assessing the fusion degree of stem cell expansion culture, the method comprising:

[0007] Acquire two-dimensional and three-dimensional images, microenvironment parameters, and stem cell fusion parameters of the environment for stem cell expansion culture at multiple time points;

[0008] The microenvironment characteristics at each time step were calculated using microenvironment parameters and stem cell fusion parameters from all time steps.

[0009] Time perception is performed based on the two-dimensional images at all times to obtain the time perception label for each time. The fusion degree of the two-dimensional images at each time is classified to obtain the fusion degree prediction label for each time.

[0010] For each time point, based on the time-aware label and fusion degree prediction label corresponding to that time point, the two-dimensional image and the three-dimensional image of that time point are fused to obtain the enhanced stem cell image;

[0011] For each enhanced stem cell image, time perception is performed to obtain an enhanced time perception label, and the fusion degree of the enhanced stem cell image is evaluated to obtain an enhanced fusion degree prediction label.

[0012] Based on all microenvironment features, all time-aware labels, all fusion degree prediction labels, all enhanced time-aware labels, and all enhanced fusion degree prediction labels, the fusion degree evaluation model is used to evaluate the fusion degree and obtain the fusion degree evaluation value at each time point.

[0013] The fusion evaluation model is trained using all fusion evaluation values ​​to obtain the trained fusion evaluation model. The trained fusion evaluation model is then used to evaluate the fusion of the environment at the time to be evaluated, and the final fusion evaluation value is obtained.

[0014] Optional, microenvironment parameters include environmental parameters, nutrient and physicochemical parameters, and genetic material enrichment parameters;

[0015] The microenvironment characteristics at each time step were calculated using microenvironment parameters and stem cell fusion parameters from all time steps, including:

[0016] Through the formula:

[0017] Data mv (t)=Mantel_test(P1(t)∪P2(t)∪P3(t),D(t-Δt)∪D(t)∪D(t+Δt), stp≤p 阈值2)

[0018] Calculate the microenvironment characteristics Data at time t. mv (t);

[0019] Where, Mantel_test() represents the Mantel test, P1(t) represents the environmental parameter at time t, P2(t) represents the nutrient and physicochemical parameters at time t, P3(t) represents the genotypic enrichment parameter at time t, D(t) represents the stem cell fusion parameter at time t, D(t-Δt) represents the stem cell fusion parameter at time t-Δt, D(t+Δt) represents the stem cell fusion parameter at time t+Δt, t∈{1,2,,...,T}, T represents the last time in the cell expansion culture cycle, st represents constraints, p represents significance, and p 阈值2 Indicates the significance threshold.

[0020] Optionally, time perception is performed based on the two-dimensional images at all times to obtain a time-perceived label for each time moment, including:

[0021] For each time point, perform the following steps:

[0022] The two-dimensional image at any given time is cropped into multiple image blocks;

[0023] The time-aware model is used to extract features from each image block to obtain the image features of each image block. Based on all image features, the time-aware label of each image block is calculated. The time-aware labels of all image blocks are aggregated to obtain the initial time-aware label of each time.

[0024] A time-aware loss function is constructed based on the initial time-aware labels, and the time-aware loss function is minimized based on the time-aware model to obtain the time-aware labels at each time step.

[0025] Optionally, the time-aware labels of all image patches are aggregated to obtain the initial time-aware labels for each moment, including:

[0026] Through the formula:

[0027]

[0028] Calculate the initial time-sensing label T′(t) at time t;

[0029] Where Pooling_aggr() represents pooling aggregation, and K represents the number of hidden layers. This represents the weighting coefficient of the l-th hidden layer at time t. Let l represent the feature representation of the l-th hidden layer corresponding to the temporal-aware labels of all image patches at time t. This represents the message construction function for the l-th hidden layer at time t. This represents the permutation-invariant aggregation function of the l-th hidden layer at time t. This represents the update function of the l-th hidden layer at time t. R represents the embedded feature representation after pooling the node features of the last hidden layer at time t. 1×1 Indicates dimension;

[0030] The time-aware loss function is:

[0031] min(L whole-time-aware (t))=min(λ·L patch-time-aware (t)+(1-λ)·L patchs-time-aggr (t))

[0032] Among them, L whole-time-aware (t) represents the value of the time-aware loss function, λ represents the weighting coefficient, and L patch-time-aware (t) represents the feature loss function, L patchs-time-aggr (t) represents the aggregation loss function:

[0033] L patch-time-aware (t)=α·L features (p (t,t±Δt) )+β·L remove_corr (s (t,t±Δt) )+γ·L errors (y (t,t±Δt) ≠y ( ' t,t±Δt) )

[0034]

[0035] L patchs-time-aggr (t)=Probability(T(t)≠T'(t))

[0036] Where α, β, and γ all represent weighting coefficients, L features () represents the feature loss function, p (t,t±Δt) L represents all image patches corresponding to time t, time t-Δt, and time t+Δt. remove_corr () represents the loss function for removing relevance from stacked features, s (t,t±Δt) L represents all image features corresponding to time t, time t-Δt, and time t+Δt. errors () represents the error rate loss function, y (t,t±Δt)Let y′ represent the temporal-aware labels of all image patches corresponding to time t, time t-Δt, and time t+Δt. (t,t±Δt) Let C represent the temporal-aware true labels for all image patches corresponding to time t, time t-Δt, and time t+Δt, where δ represents the weighting coefficient. (i,i) C represents the element in the i-th row and i-th column of the correlation matrix. (i,j) Let represent the element in the i-th row and j-th column of the correlation matrix, Probability() represents the probability, T(t) represents the time-aware true label at time t, t∈{1,2,,...,T}, and T represents the last time in the cell expansion culture cycle.

[0037] T(t) represents the time-aware real label at time t, where t∈{1,2,,...,T} and T represents the last moment in the cell expansion culture cycle.

[0038] Optionally, the fusion degree of the two-dimensional image at each time step is classified to obtain the fusion degree prediction label at each time step, including:

[0039] For each time point, perform the following steps:

[0040] The two-dimensional image at any given time is cropped into multiple image blocks;

[0041] The fusion degree classification model is used to classify the fusion degree of each image patch to obtain the image patch fusion degree prediction label. The fusion degree prediction labels of all image patches are aggregated to obtain the initial fusion degree prediction label at time step.

[0042] Construct a fusion loss function based on the predicted labels from the initial fusion degree;

[0043] Based on the fusion degree classification model, the fusion degree loss function is minimized to obtain the fusion degree prediction label at time step.

[0044] Optionally, the fusion loss function is:

[0045] min(L whole-2D-degree (t))=min(ε·{L errors (d (t) ≠d′ (t) )|stL features (p (t) )}+(1-ε)·L patchs-degree-aggr (t))

[0046] Among them, L whole-2D-degree (t) represents the value of the fusion loss function, ε represents the weight coefficient, and L errors () represents the error rate loss function, d (t)d′ represents the true label of the fusion degree of all image patches. (t) The label represents the predicted fusion degree of all image patches, st represents the constraint, and L represents the fusion degree prediction label. features () represents the feature loss function, p (t) L represents all image patches at time t. patchs-degree-aggr (t) represents the aggregation loss function for fusion degree:

[0047] L patchs-degree-aggr (t)=Probability(D(t)≠D'(t))

[0048] Where Probability() represents probability, D (t) The true label representing the degree of blending is D′. (t) This indicates the initial fusion prediction label.

[0049] Optionally, based on the time-aware labels and fusion degree prediction labels corresponding to each time moment, the two-dimensional and three-dimensional images at each time moment are fused to obtain enhanced stem cell images, including:

[0050] Through the formula:

[0051]

[0052] Acquiring enhanced stem cell images

[0053] in, Indicates an enhanced 2D image. Representing a two-dimensional projected image:

[0054]

[0055]

[0056]

[0057] Among them, w 21 w 22 w 23 w 24 All represent weights. This represents the image obtained by performing a geometric transformation on a two-dimensional image. This refers to an image obtained by using temporally aware labels and fusion prediction labels as priors, and then performing variational autoencoder enhancement on a two-dimensional image. This refers to the image obtained by using temporally aware labels and fusion prediction labels as priors, and then enhancing the 2D image with a generative adversarial network. This represents the image obtained by using temporally aware labels and fusion prediction labels as priors to perform a two-dimensional diffusion model enhancement on a two-dimensional image. `views` represents the viewpoints of the image enhancement, and `Views` indicates the preset number of multiple viewpoints for image enhancement. `Projectionoperator` represents the projection operator. Represents a three-dimensional image. Represents a three-dimensional diffusion model. This represents the generation function of the three-dimensional diffusion model. 'st' represents a two-dimensional image, and 'st' represents a constraint.

[0058] Optionally, based on all microenvironment features, all time-aware labels, all fusion degree prediction labels, all enhanced time-aware labels, and all enhanced fusion degree prediction labels, a fusion degree evaluation model is used to evaluate the fusion degree, obtaining the fusion degree evaluation value at each time step, including:

[0059] For each time point, perform the following steps:

[0060] Multi-scale feature fusion data is calculated based on the micro-environment features corresponding to each time moment, time-aware labels, fusion degree prediction labels, enhanced time-aware labels, and enhanced fusion degree prediction labels.

[0061] The fusion degree evaluation model is used to evaluate the fusion degree of multi-scale feature fusion data and obtain the fusion degree evaluation value at each time step.

[0062] Optionally, multi-scale feature fusion data is calculated based on the micro-environment features corresponding to the time, time-aware labels, fusion degree prediction labels, enhanced time-aware labels, and enhanced fusion degree prediction labels, including:

[0063] Through the formula:

[0064] Data fusion (t)=[Data mv (t),Data ota (t)∪Data ata (t),Data od (t)∪Data ad (t)]

[0065] Computing multi-scale feature fusion data fusion (t);

[0066] Among them, Data mv (t) represents the microenvironment characteristics at time t, Data ota (t) represents the time-aware label corresponding to the t-th time, Data ata (t) represents the enhanced time-aware label corresponding to the t-th time, Data od(t) represents the predicted fusion degree label at time t, Data ad (t) represents the enhanced fusion prediction label corresponding to the t-th time, t∈{1,2,,...,T}, where T represents the last time in the cell expansion culture cycle.

[0067] Secondly, this application provides a device for evaluating the fusion degree of stem cell expansion culture, comprising:

[0068] The acquisition module acquires two-dimensional and three-dimensional images, microenvironment parameters, and stem cell fusion parameters of the environment for stem cell expansion culture at multiple time points.

[0069] The calculation module uses microenvironmental parameters and stem cell fusion parameters from all time points to calculate the microenvironmental characteristics at each time point;

[0070] The first-time perception module performs time perception based on the two-dimensional images at all times, obtains the time perception label for each time, and performs fusion degree classification on the two-dimensional images at each time to obtain the fusion degree prediction label for each time.

[0071] The fusion module, for each time moment, fuses the two-dimensional and three-dimensional images based on the time-aware label and fusion degree prediction label corresponding to that time moment to obtain the enhanced stem cell image;

[0072] The second time-aware module performs time-awareness on each enhanced stem cell image to obtain an enhanced time-aware label, and evaluates the fusion degree of the enhanced stem cell image to obtain an enhanced fusion degree prediction label.

[0073] The fusion assessment module, based on all micro-environment features, all time-aware labels, all fusion prediction labels, all enhanced time-aware labels, and all enhanced fusion prediction labels, uses the fusion assessment model to assess the fusion degree and obtain the fusion assessment value at each time point.

[0074] The training module uses all the fusion evaluation values ​​to train the fusion evaluation model, obtains the trained fusion evaluation model, and uses the trained fusion evaluation model to evaluate the fusion of the environment at the time to be evaluated, and obtains the final fusion evaluation value.

[0075] Thirdly, embodiments of this application provide a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned method for evaluating the fusion degree of stem cell expansion culture.

[0076] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for evaluating the fusion degree of stem cell expansion culture.

[0077] The above-mentioned solution in this application has the following beneficial effects:

[0078] In the embodiments of this application, the microenvironmental features at each moment are calculated using microenvironmental parameters and stem cell fusion degree parameters at all moments. Then, time perception is performed based on the two-dimensional images at all moments to obtain time perception labels for each moment. The fusion degree of the two-dimensional images at each moment is classified to obtain fusion degree prediction labels for each moment. Then, for each moment, based on the time perception labels and fusion degree prediction labels corresponding to the moment, the two-dimensional and three-dimensional images at that moment are fused to obtain enhanced stem cell images. Then, for each enhanced stem cell image, time perception is performed to obtain enhanced time perception labels. The fusion degree of the enhanced stem cell images is evaluated to obtain enhanced fusion degree prediction labels. Then, based on all microenvironmental features, all time perception labels, all fusion degree prediction labels, all enhanced time perception labels, and all enhanced fusion degree prediction labels, the fusion degree evaluation model is used to evaluate the fusion degree to obtain the fusion degree evaluation value at each moment. Finally, the fusion degree evaluation model is trained using all fusion degree evaluation values ​​to obtain the trained fusion degree evaluation model. The trained fusion degree evaluation model is then used to evaluate the fusion degree of the environment at the moment to be evaluated to obtain the final fusion degree evaluation value. Among them, the fusion of two-dimensional and three-dimensional images based on time-aware labels and fusion degree prediction labels to obtain enhanced stem cell images can improve the information richness and accuracy of enhanced stem cell images. Using highly accurate enhanced stem cell images and combining them with microenvironment features for fusion degree assessment improves the information richness of fusion degree assessment and effectively enhances the accuracy of fusion degree assessment.

[0079] Other beneficial effects of this application will be described in detail in the following detailed description section. Attached Figure Description

[0080] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0081] Figure 1 is a flowchart of a method for evaluating the fusion degree of stem cell expansion culture provided in an embodiment of this application;

[0082] Figure 2 is a schematic diagram of the structure of a stem cell expansion culture fusion evaluation device provided in an embodiment of this application;

[0083] Figure 3 is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. Detailed Implementation

[0084] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0085] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0086] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0087] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0088] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0089] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0090] To address the low accuracy of existing fusion degree assessment methods, this application provides a method for assessing the fusion degree of stem cell expansion culture. This method calculates the microenvironmental characteristics at each time step using microenvironmental parameters and stem cell fusion degree parameters from all time steps. It then performs time-sensing based on two-dimensional images from all time steps to obtain a time-sensing label for each time step. Furthermore, it classifies the fusion degree of the two-dimensional images at each time step to obtain a predicted fusion degree label. Finally, for each time step, based on the corresponding time-sensing label and predicted fusion degree label, it fuses the two-dimensional and three-dimensional images to obtain an enhanced stem cell image. Finally, it performs fusion assessment on each enhanced stem cell image. Stem cell images are processed by a process involving time-aware fusion of enhanced stem cell images to obtain enhanced time-aware labels. Fusion degree assessment of these enhanced stem cell images yields enhanced fusion degree prediction labels. Then, based on all microenvironment features, time-aware labels, fusion degree prediction labels, enhanced time-aware labels, and enhanced fusion degree prediction labels, a fusion degree assessment model is used to evaluate the fusion degree at each time step, obtaining a fusion degree assessment value. Finally, the fusion degree assessment model is trained using all fusion degree assessment values ​​to obtain a trained fusion degree assessment model. This trained model is then used to evaluate the fusion degree of the environment at the time step being assessed, yielding the final fusion degree assessment value. The method of fusing 2D and 3D images based on time-aware labels and fusion degree prediction labels to obtain enhanced stem cell images improves the information richness and accuracy of the enhanced stem cell images. Using highly accurate enhanced stem cell images combined with microenvironment features for fusion degree assessment enhances the information richness of the fusion degree assessment, effectively improving its accuracy.

[0091] The following is an exemplary description of the method for evaluating the fusion degree of stem cell expansion culture provided in this application.

[0092] As shown in Figure 1, the method for evaluating the fusion degree of stem cell expansion culture provided in this application includes the following steps:

[0093] Step 11: Obtain two-dimensional images, three-dimensional images, microenvironment parameters, and stem cell fusion parameters of the environment for stem cell expansion culture at multiple time points.

[0094] The above-mentioned environment refers to the culture environment in which stem cell expansion culture is being carried out, such as a petri dish. The microenvironment parameters mentioned above include parameters related to stem cell culture, such as temperature, humidity, and hormone concentration. The stem cell fusion parameter is the ratio of the area of ​​stem cells covering the surface of the culture environment to the surface area of ​​the culture environment. Multiple moments refer to multiple historical moments with real stem cell fusion parameters.

[0095] In some embodiments of this application, two-dimensional images can be acquired using devices such as cameras, three-dimensional images can be acquired using computer three-dimensional modeling software, and microenvironment parameters can be acquired using devices such as sensors.

[0096] Step 12: Calculate the microenvironment characteristics at each time step using the microenvironment parameters and stem cell fusion parameters at all time steps.

[0097] The aforementioned microenvironment parameters include environmental parameters (such as temperature, carbon dioxide concentration, humidity, oxygen concentration, pressure, light, vibration / shaking frequency, etc.), nutrient and physicochemical parameters (such as amino acids, glucose, growth factors, vitamins, hormones, antibiotics, salts, minerals, as well as pH, osmotic pressure, etc.), and genetic material enrichment parameters (such as gene annotation enrichment, metabolic pathway enrichment, protein interaction enrichment, etc.).

[0098] Specifically, through the formula:

[0099] Data mv (t)=Mantel_test(P1(t)∪P2(t)∪P3(t),D(t-Δt)∪D(t)∪D(t+Δt), stp≤p 阈值2 )

[0100] Calculate the microenvironment characteristics Data at time t. mv (t).

[0101] Where, Mantel_test() represents the Mantel test, P1(t) represents the environmental parameter at time t, P2(t) represents the nutrient and physicochemical parameters at time t, P3(t) represents the genotypic enrichment parameter at time t, D(t) represents the stem cell fusion parameter at time t, D(t-Δt) represents the stem cell fusion parameter at time t-Δt, D(t+Δt) represents the stem cell fusion parameter at time t+Δt, t∈{1,2,,...,T}, T represents the last time in the cell expansion culture cycle, st represents constraints, p represents significance, and p 阈值2Indicates the significance threshold.

[0102] It should be noted that the formula for calculating the genetic material enrichment parameter is as follows:

[0103]

[0104] Where, E(t), E GO (t), E KEGG (t) and E PPI (t) represents the overall enrichment, Gene Ontology (GO) enrichment, Kyoto Encyclopedia of Genes and Genomes (KEGG) metabolic pathway enrichment, and Protein-Protein Interaction (PPI) enrichment results at time t, respectively; st represents the constraints, and p.adjust is the corrected p-value adjusted by the Bonferroni / Benjamini / false positive rate correction method. 阈值1 N is the preset significance threshold. GO (t), N KEGG (t), N PPI (t) represents the enriched entries for GO, KEGG, and PPI at time t; T represents the entire cell expansion and culture cycle; ∪ represents the union symbol; term represents the union of enriched entries at all times; BP, CC, and MF represent the biological processes, cellular components, and molecular functions enriched in GO, respectively; KEGG_passway represents the KEGG enrichment pathway, including metabolism, genetic information processing, environmental information processing, cellular processes, organismal systems, human diseases, and drug development; PPI_passway represents the PPI enrichment pathway, including physical binding, functional synergy, protein complex construction, regulatory networks, and pathway-level interactions.

[0105] The gene enrichment parameters are the enrichment fold that meets the above conditions, the ratio of the number of differentially expressed genes in the enriched entry to the total number of differentially expressed genes (GeneRAatio), the ratio of the total number of genes in the entry to the total number of annotated genes in human stem cells (BgRatio), the adjusted p-value (p.adjust), and the number of differentially expressed genes enriched into the entry (Count).

[0106] Step 13: Perform time perception based on the two-dimensional images at all times to obtain the time perception label for each time, and perform fusion degree classification on the two-dimensional images at each time to obtain the fusion degree prediction label for each time.

[0107] The aforementioned time-aware labels are used to describe the temporal position of a moment among all moments, while the fusion prediction labels are used to describe the degree of fusion of stem cells in the time environment corresponding to the two-dimensional image (e.g., high fusion, medium fusion, low fusion, etc.).

[0108] In some embodiments of this application, the steps of performing time perception based on two-dimensional images at all times to obtain a time-perceived label for each time moment, and performing fusion degree classification on the two-dimensional images at each time moment to obtain a fusion degree prediction label for each time moment are specifically as follows:

[0109] The first step is to perform time perception based on the two-dimensional images at all times to obtain the time perception label for each time.

[0110] For each time point, perform the following steps:

[0111] First, the two-dimensional image at each moment is cropped into multiple image blocks.

[0112] Then, the time-aware model is used to extract features from each image block to obtain the image features of each image block. Based on all image features, the image block time-aware label of each image block is calculated. The time-aware labels of all image blocks are aggregated to obtain the initial time-aware label of the time.

[0113] It should be noted that image patch feature extraction can be performed using spatial and channel-based dual attention algorithms to obtain discriminative image features. The calculation formula is as follows:

[0114]

[0115] in, It is a comprehensive attention weight. It is a spatial information extraction function implemented through convolution operations and global pooling. It is a channel information extraction function implemented through global pooling and a multilayer perceptron; patch whole For the overall patch image, patch seg For the patch image used in image segmentation, L features () represents the feature loss function. These represent feature maps at different depths and scales, where depth indicates the number of layers in the feature map and scale indicates the size of the map. It represents the combined attention weights, Loss represents the loss, and sigmoid represents the activation function.

[0116] A graph neural network can be used to calculate the temporal label of each image patch. The update formula for each hidden layer in the graph neural network is as follows:

[0117]

[0118] Among them, (X) t A t ) represents the constructed graph, X t ∈R m×1 This is the feature matrix of the image features, where m is the number of GCN nodes, and its value is the number of patches; 1 indicates the time-aware feature dimension; A t The adjacency matrix is ​​defined as follows: when there is an edge between two nodes (multiple nodes correspond to multiple image patches, and the edge is determined based on the geometric position, pixel features, and topological relationship between the corresponding two image patches; for example, if two image patches are directly adjacent in space, it is assumed that there is an edge between their corresponding two nodes), the value at the corresponding position in the matrix is ​​1; otherwise, it is 0. l is the l-th hidden layer of the GCN; Con is the message construction function; Per is the permutation-invariant aggregation function, which aggregates all messages passed to a node; Up is the update function that updates the features of existing nodes; V is a node in the graph; Mes (t,v) This is the aggregation of all messages at node V.

[0119] The above calculation of image patch time-aware labels for each image patch based on all image features, and aggregation of all image patch time-aware labels, yields the initial time-aware label for each moment:

[0120] Through the formula:

[0121]

[0122] Calculate the initial time-sensing label T′(t) at time t;

[0123] Where Pooling_aggr() represents pooling aggregation, and K represents the number of hidden layers. Represents the weighting coefficient of the l-th hidden layer. Let l represent the feature representation of the l-th hidden layer corresponding to the temporal-aware labels of all image patches at time t. This represents the message construction function for the l-th hidden layer at time t. This represents the permutation-invariant aggregation function of the l-th hidden layer at time t. This represents the update function of the l-th hidden layer at time t. R represents the embedded feature representation after pooling the node features of the last hidden layer at time t. 1×1 Indicates dimension.

[0124] It is understandable that the operations in the time-aware model are the aforementioned operations of the dual attention algorithm based on space and channel, graph neural network operations, and the aforementioned operations for calculating the initial time-aware labels.

[0125] Finally, a time-aware loss function is constructed based on the initial time-aware labels, and the time-aware loss function is minimized based on the time-aware model to obtain the time-aware labels at each time step.

[0126] Specifically, the time-aware loss function is:

[0127] min(L whole-time-aware (t))=min(λ·L patch-time-aware (t)+(1-λ)·L patchs-time-aggr (t))

[0128] Among them, L whole-time-aware (t) represents the value of the time-aware loss function, λ represents the weighting coefficient, and L patch-time-aware (t) represents the feature loss function, L patchs-time-aggr (t) represents the aggregation loss function:

[0129] L patch-time-aware (t)=α·L features (p (t,t±Δt) )+β·L remove_corr (s (t,t±Δt) )+γ·L errors (y (t,t±Δt) ≠y ( ' t,t±Δt) )

[0130]

[0131] L patchs-time-aggr (t)=Probability(T(t)≠T'(t))

[0132] Where α, β, and γ all represent weighting coefficients, L features () represents the feature loss function, p (t,t±Δt) L represents all image patches corresponding to time t, time t-Δt, and time t+Δt. remove_corr () represents the loss function for removing relevance from stacked features, s (t,t±Δt) L represents all image features corresponding to time t, time t-Δt, and time t+Δt. errors () represents the error rate loss function, y (t,t±Δt) Let y′ represent the temporal-aware labels of all image patches corresponding to time t, time t-Δt, and time t+Δt. (t,t±Δt)Let C represent the temporal-aware true labels for all image patches corresponding to time t, time t-Δt, and time t+Δt, where δ represents the weighting coefficient. (i,i) C represents the element in the i-th row and i-th column of the correlation matrix. (i,j) Let represent the element in the i-th row and j-th column of the correlation matrix, Probability() represents the probability, T(t) represents the time-aware true label at time t, t∈{1,2,,...,T}, and T represents the last time in the cell expansion culture cycle.

[0133] For example, algorithms such as gradient descent can be used to minimize the loss function. When the time-aware loss function is not minimized, the parameters of the time-aware model are adjusted, and the process of using the time-aware model to extract features from each image patch to obtain the image features of each image patch is repeated. Based on all image features, the image patch time-aware label of each image patch is calculated, and all image patch time-aware labels are aggregated to obtain the initial time-aware label at time step 1. This process continues until the loss function is minimized. At this point, the time-aware model is used to extract features from each image patch to obtain the image features of each image patch. Based on all image features, the image patch time-aware label of each image patch is calculated, and all image patch time-aware labels are aggregated to obtain the time-aware label at time step 2.

[0134] The second step is to classify the fusion degree of the two-dimensional image at each time step to obtain the fusion degree prediction label at each time step.

[0135] For each time point, perform the following steps:

[0136] First, the two-dimensional image at each moment is cropped into multiple image blocks.

[0137] Then, the fusion degree classification model is used to classify the fusion degree of each image patch to obtain the image patch fusion degree prediction label, and all image patch fusion degree prediction labels are aggregated to obtain the initial fusion degree prediction label at time step.

[0138] It should be noted that the dual attention algorithm based on space and channel mentioned above can be used to extract features from each image patch. Then, a graph convolutional network can be used to classify the fusion degree of each image patch to obtain the image patch fusion degree prediction label. Finally, global attention can be used to aggregate all the image patch fusion degree prediction labels to obtain the initial fusion degree prediction label at time step.

[0139] Understandably, the fusion classification model operates by employing a dual attention algorithm based on spatial and channel-based approaches, a graph convolutional neural network, and global attention, in that order.

[0140] Then, a fusion loss function is constructed based on the predicted labels of the initial fusion degree.

[0141] The above fusion loss function is:

[0142] min(L whole-2D-degree (t))=min(ε·{L errors (d (t) ≠d′ (t) )|stL features (p (t) )}+(1-ε)·L patchs-degree-aggr (t))

[0143] Among them, L whole-2D-degree (t) represents the value of the fusion loss function, ε represents the weight coefficient, and L errors () represents the error rate loss function, d (t) d′ represents the true label of the fusion degree of all image patches. (t) The label represents the predicted fusion degree of all image patches, st represents the constraint, and L represents the fusion degree prediction label. features () represents the feature loss function, p (t) L represents all image patches at time t. patchs-degree-aggr (t) represents the aggregation loss function for fusion degree:

[0144] L patchs-degree-aggr (t)=Probability(D(t)≠D'(t))

[0145] Among them, D (t) The true label representing the degree of blending is D′. (t) This indicates the initial fusion prediction label.

[0146] Finally, based on the fusion degree classification model, the fusion degree loss function is minimized to obtain the fusion degree prediction label at time step.

[0147] For example, algorithms such as gradient descent can be used to minimize the loss function. When the loss function is not minimized, the parameters of the fusion classification model are adjusted, and the process is repeated to classify each image patch by fusion degree using the fusion classification model, obtain the image patch fusion degree prediction label, and aggregate all the image patch fusion degree prediction labels to obtain the initial fusion degree prediction label at time step. This process continues until the loss function is minimized, and the fusion classification model at this time is used to classify each image patch by fusion degree, obtain the image patch fusion degree prediction label, and aggregate all the image patch fusion degree prediction labels to obtain the fusion degree prediction label at time step.

[0148] It should be noted that since both the time-aware model and the fusion classification model have a dual attention algorithm based on space and channel for feature extraction of image patches, the parameters of the dual attention algorithm based on space and channel in the time-aware model have already been trained in the relevant steps of the time-aware model. In this step, the parameters of the dual attention algorithm based on space and channel in the fusion classification model do not need to be trained. The parameters of the dual attention algorithm based on space and channel in the time-aware model after training can be used directly to improve the model training efficiency.

[0149] Step 14: For each time point, based on the time-aware label and fusion degree prediction label corresponding to that time point, fuse the two-dimensional image and the three-dimensional image of that time point to obtain the enhanced stem cell image.

[0150] Specifically, through the formula:

[0151]

[0152] Acquiring enhanced stem cell images

[0153] in, Indicates an enhanced 2D image. Representing a two-dimensional projected image:

[0154]

[0155]

[0156]

[0157] Among them, w 21 w 22 w 23 w 24 All represent weights. This represents the image obtained by performing a geometric transformation on a two-dimensional image. This refers to an image obtained by using temporally aware labels and fusion prediction labels as priors, and then performing variational autoencoder enhancement on a two-dimensional image. This refers to the image obtained by using temporally aware labels and fusion prediction labels as priors, and then enhancing the 2D image with a generative adversarial network. This represents the image obtained by using temporally aware labels and fusion prediction labels as priors to perform a two-dimensional diffusion model enhancement on a two-dimensional image. `views` represents the viewpoints of the image enhancement, and `Views` indicates the preset number of multiple viewpoints for image enhancement. `Projectionoperator` represents the projection operator. Represents a three-dimensional image. Represents a three-dimensional diffusion model. This represents the generation function of the three-dimensional diffusion model. 'st' represents a two-dimensional image, and 'st' represents a constraint.

[0158] Step 15: For each enhanced stem cell image, perform time perception on the enhanced stem cell image to obtain an enhanced time perception label, and evaluate the fusion degree of the enhanced stem cell image to obtain an enhanced fusion degree prediction label.

[0159] Specifically, the steps for performing time-aware analysis on enhanced stem cell images to obtain enhanced time-aware labels are the same as those in step 13 for performing time-aware analysis on two-dimensional images at all times to obtain time-aware labels for each time moment. Since the time-aware model has already been trained in step 13, the enhanced time-aware labels can be directly calculated using the trained time-aware model in this step. That is, the enhanced stem cell image is cropped into multiple image blocks; the trained time-aware model is used to extract features from each image block to obtain the image features of each image block; the image block time-aware label is calculated based on all image features; and all image block time-aware labels are aggregated to obtain the enhanced time-aware label.

[0160] The step of evaluating the fusion degree of the enhanced stem cell image to obtain the enhanced fusion degree prediction label is the same as the step of performing time-awareness based on the two-dimensional images at all times to obtain the time-aware label for each time moment in step 13. Since the parameters of the fusion degree classification model have been trained in step 13, the enhanced fusion degree prediction label can be directly calculated using the trained fusion degree classification model in this step. That is, the enhanced stem cell image is cropped into multiple image blocks; the fusion degree of each image block is classified using the trained fusion degree classification model to obtain the image block fusion degree prediction label; and all image block fusion degree prediction labels are aggregated to obtain the enhanced fusion degree prediction label.

[0161] Step 16: Based on all microenvironment features, all time-aware labels, all fusion degree prediction labels, all enhanced time-aware labels, and all enhanced fusion degree prediction labels, use the fusion degree evaluation model to evaluate the fusion degree and obtain the fusion degree evaluation value at each time point.

[0162] The above fusion assessment values ​​are stem cell fusion parameters obtained using the fusion assessment model.

[0163] For each time point, perform the following steps:

[0164] The first step is to calculate multi-scale feature fusion data based on the micro-environment features corresponding to the time, time-aware labels, fusion degree prediction labels, enhanced time-aware labels, and enhanced fusion degree prediction labels.

[0165] Specifically, through the formula:

[0166] Data fusion (t)=[Data mv (t),Data ota (t)∪Data ata (t),Data od (t)∪Data ad (t)]

[0167] Computing multi-scale feature fusion data fusion (t).

[0168] Among them, Data mv (t) represents the microenvironment characteristics at time t, Data ota (t) represents the time-aware label corresponding to the t-th time, Data ata (t) represents the enhanced time-aware label corresponding to the t-th time, Data od (t) represents the predicted fusion degree label at time t, Data ad (t) represents the enhanced fusion prediction label corresponding to the t-th time, t∈{1,2,,...,T}, where T represents the last time in the cell expansion culture cycle.

[0169] The second step is to use a fusion degree evaluation model to evaluate the fusion degree of multi-scale feature fusion data and obtain the fusion degree evaluation value at each time step.

[0170] Specifically, multi-scale feature fusion data is input into the fusion degree evaluation model to obtain the fusion degree evaluation value.

[0171] It should be noted that the above-mentioned fusion evaluation model can be a multilayer perceptron or similar model.

[0172] Step 17: Train the fusion evaluation model using all fusion evaluation values ​​to obtain the trained fusion evaluation model, and use the trained fusion evaluation model to evaluate the fusion of the environment at the time to be evaluated to obtain the final fusion evaluation value.

[0173] The final fusion evaluation value mentioned above is the stem cell fusion parameter obtained by evaluating the fusion evaluation model after training.

[0174] It should be noted that by using the fusion evaluation values ​​corresponding to all two-dimensional images and the stem cell fusion parameters as training data, the optimal model parameters θ can be obtained by performing local and global searches on the model parameters of the fusion evaluation model using an adaptive quantum evolutionary optimization algorithm. * θ * The update formula is:

[0175]

[0176] Where Loss represents the loss function (constructed based on the fusion degree assessment value and stem cell fusion degree parameters), and the initial parameters of the model are θ. This represents the fusion degree evaluation value calculated based on multi-scale feature fusion data, and argmin represents the average minimization.

[0177] For example, the steps for performing local and global searches of model parameters using the adaptive quantum evolutionary optimization algorithm are as follows:

[0178] A population of qubits is initialized using a quantum evolution algorithm, and a quantum rotation gate operation is applied to the qubit population to traverse different parameter combinations by adjusting the rotation angle.

[0179] After multiple iterations, the quantum evolutionary algorithm searches for a relatively good parameter population. From this population, a parameter combination with high fitness is selected and used as the initial point for the adaptive gradient descent algorithm.

[0180] Calculate the gradient of the loss function with respect to the model parameters, and dynamically adjust the learning rate based on the magnitude of the gradient and historical information;

[0181] Update the model parameters using the adjusted learning rate and the calculated gradient;

[0182] Repeat the above steps for multiple iterations of local optimization until the stopping condition is met.

[0183] The output layer expression of the trained fusion evaluation model after training is as follows:

[0184]

[0185] Where Activation is an activation function, the time sliding window is set to [t-Δt, t+Δt], and η t It is attention weight. This is the quantum hidden state of a multilayer quantum long short-term memory network at time t, which is updated through quantum gate operations (obtained by calculating multi-scale feature fusion data, expressed as follows). in, (for quantum gate operations); and These are the output weight matrix and bias term at time t, respectively.

[0186] The obtained optimal model parameters are substituted into the fusion evaluation model to obtain the trained fusion evaluation model. Then, based on the two-dimensional and three-dimensional images and micro-environment parameters of the environment at the time to be evaluated and the time Δt before and the time Δt after the time to be evaluated, multiple micro-environment features, multiple time-aware labels, multiple fusion prediction labels, multiple enhanced time-aware labels, and multiple enhanced fusion prediction labels are obtained according to the above steps. The trained fusion evaluation model is used to evaluate the fusion of the above data to obtain the final fusion evaluation value at the time to be evaluated.

[0187] It should be noted that in this step, since there is no corresponding stem cell fusion parameter at the time to be evaluated, blank data will be used instead of the stem cell fusion parameter in the original process when calculating the corresponding data. When performing time perception and fusion evaluation, the trained time perception model and fusion classification model will be used directly.

[0188] It is worth mentioning that fusing two-dimensional and three-dimensional images based on time-aware labels and fusion degree prediction labels to obtain enhanced stem cell images can improve the information richness and accuracy of enhanced stem cell images. Using highly accurate enhanced stem cell images and combining them with microenvironment features for fusion degree assessment improves the information richness of fusion degree assessment and effectively enhances the accuracy of fusion degree assessment.

[0189] The following is an exemplary description of the stem cell expansion culture fusion assessment device provided in this application.

[0190] As shown in Figure 2, this application embodiment provides a fusion degree assessment device for stem cell expansion culture. The fusion degree assessment device 200 for stem cell expansion culture includes:

[0191] The acquisition module 201 acquires two-dimensional images, three-dimensional images, microenvironment parameters, and stem cell fusion parameters of the environment for stem cell expansion culture at multiple time points;

[0192] The calculation module 202 calculates the microenvironment characteristics at each time step using the microenvironment parameters and stem cell fusion parameters at all times.

[0193] The first-time perception module 203 performs time perception based on the two-dimensional images at all times, obtains the time perception label for each time, and performs fusion degree classification on the two-dimensional images at each time to obtain the fusion degree prediction label for each time.

[0194] The fusion module 204 fuses the two-dimensional and three-dimensional images at each time point based on the time-aware label and fusion degree prediction label corresponding to that time point to obtain an enhanced stem cell image.

[0195] The second time-aware module 205 performs time-awareness on each enhanced stem cell image to obtain an enhanced time-aware label, and evaluates the fusion degree of the enhanced stem cell image to obtain an enhanced fusion degree prediction label.

[0196] The fusion assessment module 206, based on all micro-environment features, all time-aware labels, all fusion prediction labels, all enhanced time-aware labels, and all enhanced fusion prediction labels, uses the fusion assessment model to assess the fusion degree and obtain the fusion assessment value at each time point.

[0197] Training module 207 uses all the fusion evaluation values ​​to train the fusion evaluation model, obtains the trained fusion evaluation model, and uses the trained fusion evaluation model to evaluate the fusion of the environment at the time to be evaluated, and obtains the final fusion evaluation value.

[0198] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0199] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0200] As shown in FIG3, an embodiment of this application provides a terminal device. The terminal device D10 of this embodiment includes: at least one processor D100 (only one processor is shown in FIG3), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100. When the processor D100 executes the computer program D102, it implements the steps in any of the above-described method embodiments.

[0201] Specifically, when the processor D100 executes the computer program D102, it calculates the microenvironment features at each moment by utilizing the microenvironment parameters and stem cell fusion degree parameters at all moments. Then, it performs time perception based on the two-dimensional images at all moments to obtain a time-perceived label for each moment. It then classifies the fusion degree of the two-dimensional images at each moment to obtain a fusion degree prediction label for each moment. Finally, for each moment, based on the corresponding time-perceived label and fusion degree prediction label, it fuses the two-dimensional and three-dimensional images to obtain an enhanced stem cell image. Finally, for each enhanced stem cell image, it performs... This process involves performing time-aware fusion to obtain enhanced time-aware labels, and then assessing the fusion degree of enhanced stem cell images to obtain enhanced fusion degree prediction labels. Based on all microenvironment features, all time-aware labels, all fusion degree prediction labels, all enhanced time-aware labels, and all enhanced fusion degree prediction labels, a fusion degree assessment model is used to evaluate the fusion degree at each time step, obtaining the fusion degree assessment value. Finally, the fusion degree assessment model is trained using all fusion degree assessment values ​​to obtain a trained fusion degree assessment model. This trained model is then used to evaluate the fusion degree of the environment at the time step to be evaluated, yielding the final fusion degree assessment value. The method of fusing 2D and 3D images based on time-aware labels and fusion degree prediction labels to obtain enhanced stem cell images improves the information richness and accuracy of the enhanced stem cell images. Using highly accurate enhanced stem cell images combined with microenvironment features for fusion degree assessment enhances the information richness of the fusion degree assessment, effectively improving its accuracy.

[0202] The processor D100 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0203] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may be an external storage device of the terminal device D10, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device D10. Furthermore, the memory D101 may include both internal and external storage units of the terminal device D10. The memory D101 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory D101 can also be used to temporarily store data that has been output or will be output.

[0204] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0205] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0206] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to the apparatus / terminal device for the fusion assessment method of stem cell expansion culture, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0207] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0208] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0209] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for evaluating the fusion degree of stem cell expansion culture, characterized in that, include: Two-dimensional and three-dimensional images, microenvironment parameters, and stem cell fusion parameters of the environment for stem cell expansion culture at multiple time points are acquired; the microenvironment characteristics at each time point are calculated using the microenvironment parameters and stem cell fusion parameters at all time points. Time perception is performed based on the two-dimensional images at all times to obtain a time perception label for each time moment, and the fusion degree of the two-dimensional images at each time moment is classified to obtain a fusion degree prediction label for each time moment. For each of the stated times, based on the time-aware label and fusion degree prediction label corresponding to the stated time, the two-dimensional image and the three-dimensional image of the stated time are fused to obtain the enhanced stem cell image; For each enhanced stem cell image, time-awareness is performed to obtain an enhancement time-aware label, and fusion degree is evaluated to obtain an enhancement fusion degree prediction label. Based on all microenvironment features, all time-aware labels, all fusion degree prediction labels, all enhancement time-aware labels, and all enhancement fusion degree prediction labels, a fusion degree evaluation model is used to evaluate the fusion degree, obtaining a fusion degree evaluation value for each time point. The fusion degree evaluation model is trained using all fusion degree evaluation values ​​to obtain a trained fusion degree evaluation model, and the trained fusion degree evaluation model is used to evaluate the fusion degree of the environment at the time to be evaluated, obtaining a final fusion degree evaluation value. The microenvironment parameters include environmental parameters, nutrient and physicochemical parameters, and gene enrichment parameters. The calculation of the microenvironment features at each time point using the microenvironment parameters and stem cell fusion degree parameters at all times includes: using the formula: Calculate the first Microenvironment characteristics at a given moment ;in, This indicates the Mantel test. Indicates the first Environmental parameters at a given time, Indicates the first Nutrient and physicochemical parameters at a given time point Indicates the first Genetic material enrichment parameters at a given time point Indicates the first Stem cell fusion parameters at a given time point. Indicates the first Stem cell fusion parameters at a given time point. Indicates the first Stem cell fusion parameters at a given time point. ， This indicates the last moment in the cell expansion culture cycle. Indicates constraints. Indicates significance. The saliency threshold is used to represent the time perception. The step of performing time perception based on the two-dimensional images at all times to obtain a time-perception label for each time moment includes: for each time moment, performing the following steps: cropping the two-dimensional image at that time moment into multiple image blocks; extracting features from each image block using a time-perception model to obtain image features for each image block; calculating an image block time-perception label for each image block based on all image features; aggregating all image block time-perception labels to obtain an initial time-perception label for that time moment; constructing a time-perception loss function based on the initial time-perception label; and minimizing the time-perception loss function based on the time-perception model to obtain the time-perception label for that time moment.

2. The fusion degree evaluation method according to claim 1, characterized in that, The aggregation of time-aware labels for all image patches to obtain the initial time-aware label for the given time moment includes: using the formula: Calculate the first Initial time-aware label at each moment ;in, Indicates pooling aggregation, Indicates the number of hidden layers. Indicates the first At the [time]th moment The weighting coefficients of each hidden layer Indicates the first The time-aware labels corresponding to all image patches at time step n are the first The feature representation of each hidden layer Indicates the first The first moment mentioned A hidden layer message construction function, Indicates the first The first moment mentioned An aggregation function that is invariant to the arrangement of hidden layers. Indicates the first The first moment mentioned The update function for each hidden layer Indicates the first The embedded feature representation is obtained by pooling the node features of the last hidden layer at each time step. The dimension is represented; the time-aware loss function is: in, This represents the value of the time-aware loss function. Indicates the weighting coefficient. Represents the feature loss function, Representing the aggregation loss function: Where α, β, and γ all represent weighting coefficients. Represents the feature loss function, Indicates the first The moment, the first The moment and the All image patches corresponding to each time point The loss function represents the removal of relevance from stacked features. Indicates the first The moment, the first The moment, the first All image features corresponding to each time point Represents the error rate loss function. Indicates the first The moment, the first The moment and the Temporal-aware labels for all image patches corresponding to each time point. Indicates the first The moment, the first The moment and the The temporal-aware ground truth labels for all image patches corresponding to a given time point, where δ represents the weighting coefficient. Represents the first in the correlation matrix Line number Column elements, Represents the first in the correlation matrix Line number Column elements, Represents probability. Indicates the first Real-time perception labels at any given moment ， This indicates the last moment in the cell expansion culture cycle.

3. The fusion degree evaluation method according to claim 1, characterized in that, The step of classifying the fusion degree of the two-dimensional image at each time moment to obtain a fusion degree prediction label for each time moment includes: for each time moment, performing the following steps: cropping the two-dimensional image at that time moment into multiple image blocks; classifying the fusion degree of each image block using a fusion degree classification model to obtain an image block fusion degree prediction label, and aggregating all image block fusion degree prediction labels to obtain an initial fusion degree prediction label for that time moment; constructing a fusion degree loss function based on the initial fusion degree prediction label; and minimizing the fusion degree loss function based on the fusion degree classification model to obtain the fusion degree prediction label for that time moment.

4. The fusion degree evaluation method according to claim 3, characterized in that, The fusion loss function is: in, This represents the value of the fusion loss function. Indicates the weighting coefficient. Represents the error rate loss function. Indicates the true label of the fusion degree of all image patches. Indicates the predicted fusion degree labels for all image patches. Indicates constraints. Represents the feature loss function, Indicates the first All image patches at each time point The aggregation loss function represents the degree of fusion: in, Represents probability. Indicates the degree of integration as a true label. This indicates the initial fusion prediction label.

5. The fusion degree evaluation method according to claim 1, characterized in that, The process of fusing the two-dimensional and three-dimensional images at the specified time point, based on the time-aware label and fusion degree prediction label, to obtain the enhanced stem cell image includes: using the formula: Acquiring enhanced stem cell images ;in, Indicates an enhanced 2D image. Representing a two-dimensional projected image: in, 、 、 、 All represent weights. This represents the image obtained by performing a geometric transformation on a two-dimensional image. This indicates an image obtained by performing variational autoencoder enhancement on the two-dimensional image using temporally aware labels and fusion prediction labels as priors. This indicates that the image is obtained by using time-aware labels and fusion prediction labels as priors, and then enhancing the two-dimensional image with a generative adversarial network. This indicates the image obtained by using time-aware labels and fusion prediction labels as priors, and then performing two-dimensional diffusion model enhancement on the two-dimensional image. Indicates the viewpoint for image enhancement. This indicates the number of multi-view presets for image enhancement. This indicates the projection operator. Represents a three-dimensional image. Represents a three-dimensional diffusion model. This represents the generation function of the three-dimensional diffusion model. Representing a two-dimensional image, Indicates constraints.

6. The fusion degree evaluation method according to claim 1, characterized in that, The method of evaluating the fusion degree based on all microenvironment features, all time-aware labels, all fusion degree prediction labels, all enhanced time-aware labels, and all enhanced fusion degree prediction labels, and obtaining the fusion degree evaluation value for each time moment, includes the following steps for each time moment: calculating multi-scale feature fusion data based on the microenvironment features, time-aware labels, fusion degree prediction labels, enhanced time-aware labels, and enhanced fusion degree prediction labels corresponding to the time moment; and evaluating the fusion degree of the multi-scale feature fusion data using the fusion degree evaluation model to obtain the fusion degree evaluation value for the time moment.

7. The fusion degree evaluation method according to claim 6, characterized in that, The calculation of multi-scale feature fusion data based on the micro-environment features, time-aware labels, fusion degree prediction labels, enhanced time-aware labels, and enhanced fusion degree prediction labels corresponding to the time moment includes: using the formula: Computing multi-scale feature fusion data ;in, Indicates the first The microenvironment characteristics corresponding to each moment Indicates the first The time-aware label corresponding to each moment. Indicates the first The enhanced time-aware label corresponding to each moment. Indicates the first The predicted fusion degree label for each time point Indicates the first The predicted label for enhanced fusion at each time point ， This indicates the last moment in the cell expansion culture cycle.

8. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for evaluating the fusion degree of stem cell expansion culture as described in any one of claims 1 to 7.

Images