A method and system for predicting stem cell differentiation based on machine learning

Abstract

The present invention discloses a method and system for predicting stem cell differentiation based on machine learning, and the method includes data collection, basic differentiation classification, cell region extraction, cell image segmentation and stem cell differentiation prediction. The present invention relates to the field of stem cell differentiation prediction technology, and specifically refers to a method and system for predicting stem cell differentiation based on machine learning. The present invention constructs nine differentiation categories, and uses a convolutional neural network structure with a deepened convolutional layer for basic differentiation classification, thereby improving the versatility of the method; uses a method based on a pre-trained densely connected convolutional network for cell region extraction, thereby optimizing the overall accuracy of the method; uses a U-shaped network for cell image segmentation, thereby optimizing the recognition effect of specific categories of cell regions; uses a spatiotemporal recurrent neural network for continuous frame stem cell differentiation prediction, thereby improving the overall usability of the method and broadening the scope of application of the method.

Description

Technical Field

[0001] The present invention relates to the technical field of stem cell differentiation prediction, and specifically to a stem cell differentiation prediction method and system based on machine learning. Background Art

[0002] The stem cell differentiation prediction method based on machine learning refers to the use of machine learning algorithms to predict the transformation of cell types during stem cell differentiation. The purpose of stem cell differentiation prediction is to predict the possible differentiation results given the characteristics and environmental conditions of stem cells. By predicting the differentiation of stem cells, we can better understand the function and potential of stem cells, which is helpful to develop and optimize cell therapy programs.

[0003] However, in the existing stem cell differentiation prediction methods, there is a technical problem that the classification categories of the induced pluripotent stem cell differentiation process are insufficient, which affects the subsequent classification effect; in the existing stem cell differentiation prediction methods, there is a technical problem that the classification based on a single-frame image may classify cell regions belonging to the same category into different categories; in the existing stem cell differentiation prediction methods, there is a technical problem that the cell image segmentation accuracy is low, which affects the subsequent calculation efficiency and accuracy; in the existing stem cell differentiation prediction methods, there is a technical problem that the stem cell differentiation prediction based on a single-frame image cannot effectively reflect the complete process of cell differentiation and cannot meet the prediction needs. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a stem cell differentiation prediction method and system based on machine learning. In view of the technical problem that in the existing stem cell differentiation prediction methods, the classification categories of the induced pluripotent stem cell differentiation process are insufficient, which affects the subsequent classification effect, this solution creatively constructs nine differentiation categories, and adopts a convolutional neural network structure with a deepened convolutional layer for basic differentiation classification, thereby realizing multi-category preliminary classification of cell differentiation classification, improving the versatility of the method, and optimizing the overall automation of the method; in view of the technical problem that in the existing stem cell differentiation prediction methods, the classification based on a single frame image may classify cell regions belonging to the same category into different categories, this solution creatively adopts a method based on a pre-trained densely connected convolutional network to extract cell regions, unify the cell regions of the same category, and extract the cell regions into different categories. Extraction, optimize the overall accuracy of the method; in view of the technical problem that the existing stem cell differentiation prediction methods have low accuracy of cell image segmentation, which affects the subsequent calculation efficiency and accuracy, this scheme creatively uses a U-shaped network for cell image segmentation, optimizes the recognition effect of the area of ​​specific categories of cells, and provides a good data foundation for subsequent continuous time series prediction; in view of the technical problem that the existing stem cell differentiation prediction methods have the problem that the stem cell differentiation prediction based on a single frame image cannot effectively reflect the complete process of cell differentiation and cannot meet the prediction needs, this scheme creatively uses a spatiotemporal recurrent neural network to predict the stem cell differentiation of continuous frames, and by obtaining continuous stem cell differentiation prediction data, it better reflects the process of stem cell differentiation and the cell information in the reprogramming stage, improves the overall usability of the method, and broadens the scope of application of the method.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides a method for predicting stem cell differentiation based on machine learning, the method comprising the following steps:

[0006] Step S1: data collection;

[0007] Step S2: basic differentiation classification;

[0008] Step S3: cell region extraction;

[0009] Step S4: cell image segmentation;

[0010] Step S5: Stem cell differentiation prediction.

[0011] Further, in step S1, the data acquisition is used to acquire induced pluripotent stem cell image data, specifically using an incubator-type microscope to obtain original time-lapse microscopic image data for predicting stem cell differentiation from the microscope imaging data through time-lapse acquisition, and the original time-lapse microscopic image data for predicting stem cell differentiation specifically refers to frame-by-frame image data of stem cell differentiation time series.

[0012] Further, in step S2, the basic differentiation classification is used to define differentiation categories and generate probability images, specifically, nine differentiation categories are constructed, and a convolutional neural network structure with a deeper convolutional layer is used to perform basic differentiation classification to obtain a basic probability image of cell differentiation;

[0013] The convolutional neural network structure of the deepened convolutional layer includes a classification model deepened convolutional layer, a classification model pooling layer and a classification model fully connected layer;

[0014] The deepening convolution layer includes an expanded convolution layer and a basic convolution layer, wherein the expanded convolution layer is used to extract large-scale features, and the basic convolution layer is used to perform feature filtering on the extracted features;

[0015] The classification model pooling layer is used to perform an average pooling operation;

[0016] The fully connected layer of the classification model is used to calculate the probability of cell differentiation categories, specifically calculating the classification probability of nine types of cell differentiation images;

[0017] The steps of constructing nine differentiation categories and using a convolutional neural network structure with a deeper convolutional layer to perform basic differentiation classification to obtain a basic probability image of cell differentiation include:

[0018] Step S21: Differentiation category definition, specifically, nine categories of stem cell differentiation are defined for induced pluripotent stem cell images to obtain a stem cell differentiation category template image set I B , and according to the stem cell differentiation category template image set I B , by cutting out 2000 training sample sets from the original time-lapse microscopic image data for stem cell differentiation prediction;

[0019] The stem cell differentiation category template image set I B , including background image I BG , Dispersed Single Cell Image I DSC , single cell images in colonies or non-colonies SC , floating cells in colonies or non-colonies I FC , intermediate cell image in the colony I CC , Image of cells in the early stages of differentiation of induced pluripotent stem cells I FPSC , Induced Pluripotent Stem Cell Image I PSC , Image of late differentiation of induced pluripotent stem cells I LPSC and inter-colony accumulation cell image I PBC , the calculation formula for the differentiation category definition is:

[0020] I B ={I BG ,I DSC,I SC ,I FC ,I CC ,I FPSC ,I PSC ,I LPSC ,I PBC};

[0021] In the formula, I B is a set of template images of stem cell differentiation category, I BG is a background image, I DSC is a dispersed single cell image, I SC is a single cell image in a colony or non-colony, I FC Is the image of floating cells in colonies or non-colonies, I CC is the middle cell image in the colony, I FPSC This is an image of cells in the early stages of differentiation of induced pluripotent stem cells. PSC This is an image of induced pluripotent stem cells. LPSC This is an image of the late stage of induced pluripotent stem cell differentiation. PBC It is an image of cells piled up between colonies;

[0022] Step S22: Constructing a classification model to deepen the convolutional layer, specifically constructing 6 convolutional layers, of which 3 convolutional layers are used to construct dilated convolutional layers and 3 convolutional layers are used to construct basic convolutional layers, including the following steps:

[0023] Step S221: constructing an expanded convolutional layer, specifically by sequentially constructing a convolutional layer of size 3×3, a convolutional layer of size 6×6, and a convolutional layer of size 5×5, to construct the expanded convolutional layer;

[0024] Step S222: constructing a basic convolutional layer, specifically by constructing three convolutional layers of size 3×3 to construct the basic convolutional layer;

[0025] Step S23: constructing a classification model pooling layer, specifically, constructing a pooling layer after each dilated convolutional layer to perform an average pooling operation to construct the classification model pooling layer;

[0026] Step S24: constructing a fully connected layer of the classification model, specifically constructing three fully connected layers after the last convolution layer of the basic convolution layer to construct the fully connected layer of the classification model;

[0027] Step S25: constructing a classification loss function, specifically using a cross entropy loss function, to optimize the probability distribution;

[0028] Step S26: basic differentiation classification model training, specifically, through the differentiation category definition, the classification model deepening convolution layer, the classification model pooling layer, the classification model full connection layer and the classification loss function, the basic differentiation classification model training is performed based on the training sample set to obtain the basic differentiation classification model Model CNN ;

[0029] Step S27: basic differentiation classification, specifically using the basic differentiation classification model Model CNN , perform basic differentiation classification and obtain the basic probability image of cell differentiation.

[0030] Further, in step S3, the cell region extraction is used to extract cell regions belonging to the same category, specifically by using a method based on a pre-trained densely connected convolutional network, based on the stem cell differentiation prediction original time-lapse microscopic image data and the stem cell differentiation category template image set, to perform cell region extraction to obtain cell region image data, wherein the cell region image data is used to represent images of cell regions belonging to the same category but having a probability of being classified into different categories by the basic differentiation;

[0031] The pre-trained densely connected convolutional network includes a region extraction convolutional layer, a dense block, a transition layer, a region extraction pooling layer and a region extraction fully connected layer;

[0032] The region extraction convolution layer is used to extract image features;

[0033] The dense block is used to transmit feature map information;

[0034] The transition layer is used to reduce the dimension of the feature map;

[0035] The region extraction pooling layer is used to unify the size of feature map vectors;

[0036] The region extraction fully connected layer is used for classification and region extraction;

[0037] The method based on the pre-trained densely connected convolutional network is used to extract the cell region based on the original time-lapse microscopic image data of the stem cell differentiation prediction and the stem cell differentiation category template image set to obtain the cell region image data, including:

[0038] Step S31: constructing a region extraction convolution layer, specifically using a convolution kernel of size 1×1 to extract features, and using an S-type function for activation to construct the region extraction convolution layer;

[0039] Step S32: constructing a dense block, specifically constructing a three-layer dense block, using a ReLU activation function to activate each layer, and performing a convolution operation with a convolution kernel of size 3×3 at the end of each layer to construct the dense block;

[0040] Step S33: constructing a transition layer, specifically using a convolutional layer with a size of 1×1 and an average pooling layer with a size of 2×2 to perform a feature dimensionality reduction operation to construct the transition layer;

[0041] Step S34: construct a region extraction pooling layer, specifically, add a global pooling layer at the end of the dense block to convert the output feature map of the dense block into a vector of a fixed size of 512×512;

[0042] Step S35: constructing a region extraction fully connected layer, specifically, connecting the output of the region extraction pooling layer to the fully connected layer, performing feature image classification, and using a softmax classifier to perform a cell region extraction operation to construct the region extraction fully connected layer;

[0043] Step S36: region extraction model training, specifically, by constructing the region extraction convolution layer, the dense block, the transition layer, the region extraction pooling layer and the region extraction fully connected layer, and according to the stem cell differentiation category template image set, 600 training sample images are extracted from the original time-lapse microscopy image data of stem cell differentiation prediction by random cropping, and the region extraction model training is performed to obtain the region extraction model Model DN ;

[0044] Step S37: Cell region extraction, specifically using the region extraction model Model DN , perform cell region extraction and obtain cell region image data.

[0045] Further, in step S4, the cell image segmentation is used to segment the cell image and extract cell features, specifically using a U-shaped network to perform cell image segmentation on the original time-lapse microscopic image data for stem cell differentiation prediction to obtain a segmented cell image;

[0046] The U-shaped network includes a downsampling subnet and an upsampling subnet;

[0047] The downsampling subnet is used to extract high-level abstract features;

[0048] The upsampling subnet is used to retain semantic information and perform feature upsampling and fusion;

[0049] The step of using a U-shaped network to perform cell image segmentation on the original time-lapse microscopic image data for stem cell differentiation prediction to obtain a segmented cell image comprises:

[0050] Step S41: constructing a downsampling subnet, specifically constructing four double-layer convolutional layers, including constructing a first downsampling convolutional layer, a second downsampling convolutional layer, a third downsampling convolutional layer and a fourth downsampling convolutional layer, setting a maximum pooling layer below each double-layer convolutional layer, and setting a connection layer below each maximum pooling layer, including a first connection layer, a second connection layer, a third connection layer and a fourth connection layer, the connection layer is used to connect with the convolutional layer in the upsampling subnet, perform a cell feature extraction operation, and construct the downsampling subnet;

[0051] Step S42: constructing an upsampling subnet, specifically constructing 5 double-layer convolution layers, including constructing a first upsampling convolution layer, a second upsampling convolution layer, a third upsampling convolution layer, a fourth upsampling convolution layer and a fifth upsampling convolution layer, and setting an upsampling layer below the first upsampling convolution layer, the second upsampling convolution layer, the third upsampling convolution layer and the fourth upsampling convolution layer, and setting a softmax classifier below the fifth upsampling convolution layer, the fifth upsampling convolution layer is connected to the maximum pooling layer of the first downsampling convolution layer through the first connection layer, the fourth upsampling convolution layer is connected to the maximum pooling layer of the second downsampling convolution layer through the second connection layer, the third upsampling convolution layer is connected to the maximum pooling layer of the third downsampling convolution layer through the third connection layer, and the second upsampling convolution layer is connected to the maximum pooling layer of the fourth downsampling convolution layer through the fourth connection layer, performing feature fusion operation, and constructing the upsampling subnet;

[0052] Step S43: constructing a cell segmentation model loss function, specifically comprising the following steps:

[0053] Step S431: construct a weight loss function, the calculation formula is:

[0054] ;

[0055] Where, L W is the weight loss function, N is the total number of samples, j is the sample index, and E is the cross entropy loss function. The calculation formula of the cross entropy loss function E is: E=-Y×logX, where Y is the true value of cell segmentation, X is the predicted value of cell segmentation, is the loss weight;

[0056] Step S432: construct a scatter loss function, the calculation formula is:

[0057] ;

[0058] Where, L D is the scatter loss function, X is the predicted value of cell segmentation, Y is the true value of cell segmentation, is the intersection operator, |·| is the modulus operator;

[0059] Step S433: construct a cell segmentation model loss function, the calculation formula is:

[0060] ;

[0061] Where L is the cell segmentation model loss function, is the weight loss balance coefficient, L W is the weight loss function, is the scatter loss balance coefficient, L D is the scatter loss function;

[0062] Step S44: Cell segmentation model training, specifically, by constructing the downsampling subnet, constructing the upsampling subnet and constructing the cell segmentation model loss function, the cell segmentation model is trained to obtain the cell segmentation model Model UN ;

[0063] Step S45: Cell image segmentation, specifically using the cell segmentation model Model UN , perform cell image segmentation and obtain the segmented cell image.

[0064] Further, in step S5, the stem cell differentiation prediction is used to perform stem cell differentiation prediction, specifically using a spatiotemporal recurrent neural network to perform time-series stem cell differentiation prediction based on the cell differentiation basic probability image, the cell region image data and the segmented cell image to obtain continuous stem cell differentiation prediction data;

[0065] The spatiotemporal recurrent neural network comprises a feedforward training subnet, a pre-training subnet and a basic recurrent neural network;

[0066] The feedforward training subnet is used to predict the stem cell differentiation status of the next frame of the time series image at the current moment;

[0067] The pre-trained subnet is used to generate initial parameters of a basic recurrent neural network through training;

[0068] The basic recurrent neural network is used to predict stem cell differentiation;

[0069] The step of using a spatiotemporal recurrent neural network to perform time-series stem cell differentiation prediction based on the cell differentiation basic probability image, the cell region image data and the segmented cell image to obtain stem cell differentiation prediction data includes:

[0070] Step S51: constructing a feedforward training subnet, specifically by constructing a four-layer stacked long short-term memory neural network, combining the cell region image data and the segmented cell image to perform cell recognition, and performing a stem cell differentiation prediction operation for the next frame of the time series image based on the single frame image to obtain a sub-frame differentiation prediction image, and constructing the feedforward training subnet;

[0071] Step S52: constructing a pre-trained subnet, specifically using the continuous probability image in the basic probability image of cell differentiation to construct the pre-trained subnet and obtain the initial parameters of the basic recurrent neural network;

[0072] Step S53: constructing a basic recurrent neural network, specifically, based on the initial parameters of the basic recurrent neural network and the sub-frame differentiation prediction image, iteratively learning a four-layer stacked long short-term memory neural network to obtain a continuous stem cell differentiation prediction image, and constructing the basic recurrent neural network;

[0073] Step S54: continuous stem cell differentiation prediction model training, specifically, by constructing the feedforward training subnet, constructing the pre-training subnet and constructing the basic recurrent neural network, performing continuous stem cell differentiation prediction model training to obtain a continuous stem cell differentiation prediction model Model RNN ;

[0074] Step S55: Stem cell differentiation prediction, specifically using the continuous stem cell differentiation prediction model Model RNN Perform stem cell differentiation prediction to obtain continuous stem cell differentiation prediction data.

[0075] The present invention provides a stem cell differentiation prediction system based on machine learning, comprising a data acquisition module, a basic differentiation classification module, a cell region extraction module, a cell image segmentation module and a stem cell differentiation prediction module;

[0076] The data acquisition module is used for data acquisition, and obtains original time-lapse microscopic image data for stem cell differentiation prediction through data acquisition, and sends the original time-lapse microscopic image data for stem cell differentiation prediction to the basic differentiation classification module, the cell region extraction module, the cell image segmentation module and the stem cell differentiation prediction module;

[0077] The basic differentiation classification module is used for basic differentiation classification, and obtains a stem cell differentiation category template image set and a cell differentiation basic probability image through basic differentiation classification, and sends the stem cell differentiation category template image set to the cell region extraction module, and sends the cell differentiation basic probability image to the stem cell differentiation prediction module;

[0078] The cell region extraction module is used for cell region extraction, obtains cell region image data through cell region extraction, and sends the cell region image data to the stem cell differentiation prediction module;

[0079] The cell image segmentation module is used for cell image segmentation, obtains a segmented cell image through cell image segmentation, and sends the segmented cell image to the stem cell differentiation prediction module;

[0080] The stem cell differentiation prediction module is used for stem cell differentiation prediction, and obtains continuous stem cell differentiation prediction data through stem cell differentiation prediction.

[0081] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0082] (1) In view of the technical problem that the existing stem cell differentiation prediction methods have insufficient classification categories for the differentiation process of induced pluripotent stem cells, which affects the subsequent classification effect, this scheme creatively constructs nine differentiation categories and uses a convolutional neural network structure with a deeper convolutional layer for basic differentiation classification, realizing multi-category preliminary classification of cell differentiation classification, improving the versatility of the method and optimizing the overall automation of the method;

[0083] (2) In view of the technical problem that the existing stem cell differentiation prediction methods may classify cell regions belonging to the same category into different categories based on the classification of single-frame images, this solution creatively adopts a method based on pre-trained densely connected convolutional networks to extract cell regions, extract cell regions of the same category in a unified manner, and optimize the overall accuracy of the method;

[0084] (3) In view of the low accuracy of cell image segmentation in existing stem cell differentiation prediction methods, which affects the subsequent computational efficiency and accuracy, this solution creatively uses a U-shaped network for cell image segmentation, optimizes the recognition effect of specific cell regions, and provides a good data foundation for subsequent continuous time series prediction;

[0085] (4) In view of the technical problem that the existing stem cell differentiation prediction methods based on single-frame images cannot effectively reflect the complete process of cell differentiation and cannot meet the prediction needs, this scheme creatively uses a spatiotemporal recurrent neural network to predict stem cell differentiation in continuous frames. By obtaining continuous stem cell differentiation prediction data, it can better reflect the process of stem cell differentiation and the cell information in the reprogramming stage, thereby improving the overall usability of the method and broadening the scope of application of the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1A schematic diagram of a process flow of a method for predicting stem cell differentiation based on machine learning provided by the present invention;

[0087] Figure 2 A schematic diagram of a stem cell differentiation prediction system based on machine learning provided by the present invention;

[0088] Figure 3 A schematic diagram of the process of basic differentiation and classification in step S2;

[0089] Figure 4 This is a schematic diagram of the process of cell region extraction in step S3;

[0090] Figure 5 It is a schematic diagram of the process of cell image segmentation in step S4;

[0091] Figure 6 Schematic diagram of the process for stem cell differentiation prediction in step S5.

[0092] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0093] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0094] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0095] Example 1, see Figure 1 The present invention provides a method for predicting stem cell differentiation based on machine learning, the method comprising the following steps:

[0096] Step S1: data collection;

[0097] Step S2: basic differentiation classification;

[0098] Step S3: cell region extraction;

[0099] Step S4: cell image segmentation;

[0100] Step S5: Stem cell differentiation prediction.

[0101] Example 2, see Figure 1 and Figure 2 In step S1, the data acquisition is used to acquire induced pluripotent stem cell image data, specifically using an incubator-type microscope to obtain original time-lapse microscopic image data for stem cell differentiation prediction from the microscope imaging data through time-lapse acquisition. The original time-lapse microscopic image data for stem cell differentiation prediction specifically refers to frame-by-frame image data of stem cell differentiation time series.

[0102] Example 3, see Figure 1 , Figure 2 and Figure 3 , this embodiment is based on the above embodiment. In step S2, the basic differentiation classification is used to define differentiation categories and generate probability images, specifically to construct nine differentiation categories, and use a convolutional neural network structure with a deeper convolutional layer to perform basic differentiation classification, and obtain a basic probability image of cell differentiation;

[0103] The convolutional neural network structure of the deepened convolutional layer includes a classification model deepened convolutional layer, a classification model pooling layer and a classification model fully connected layer;

[0104] The deepening convolution layer includes an expanded convolution layer and a basic convolution layer, wherein the expanded convolution layer is used to extract large-scale features, and the basic convolution layer is used to perform feature filtering on the extracted features;

[0105] The classification model pooling layer is used to perform an average pooling operation;

[0106] The fully connected layer of the classification model is used to calculate the probability of cell differentiation categories, specifically calculating the classification probability of nine types of cell differentiation images;

[0107] The steps of constructing nine differentiation categories and using a convolutional neural network structure with a deeper convolutional layer to perform basic differentiation classification to obtain a basic probability image of cell differentiation include:

[0108] Step S21: Differentiation category definition, specifically, nine categories of stem cell differentiation are defined for induced pluripotent stem cell images to obtain a stem cell differentiation category template image set I B , and according to the stem cell differentiation category template image set I B , by cutting out 2000 training sample sets from the original time-lapse microscopic image data for stem cell differentiation prediction;

[0109] The stem cell differentiation category template image set I B , including background image I BG , Dispersed Single Cell Image IDSC , single cell images in colonies or non-colonies SC , floating cells in colonies or non-colonies I FC , intermediate cell image in the colony I CC , Image of cells in the early stages of differentiation of induced pluripotent stem cells I FPSC , Induced Pluripotent Stem Cell Image I PSC , Image of late differentiation of induced pluripotent stem cells I LPSC and inter-colony accumulation cell image I PBC , the calculation formula for the differentiation category definition is:

[0110] I B ={I BG ,I DSC ,I SC ,I FC ,I CC ,I FPSC ,I PSC ,I LPSC ,I PBC};

[0111] In the formula, I B is a set of template images of stem cell differentiation category, I BG is a background image, I DSC is a dispersed single cell image, I SC is a single cell image in a colony or non-colony, I FC Is the image of floating cells in colonies or non-colonies, I CC is the middle cell image in the colony, I FPSC This is an image of cells in the early stages of differentiation of induced pluripotent stem cells. PSC This is an image of induced pluripotent stem cells. LPSC This is an image of the late stage of induced pluripotent stem cell differentiation. PBC It is an image of cells piled up between colonies;

[0112] Step S22: Constructing a classification model to deepen the convolutional layer, specifically constructing 6 convolutional layers, of which 3 convolutional layers are used to construct dilated convolutional layers and 3 convolutional layers are used to construct basic convolutional layers, including the following steps:

[0113] Step S221: constructing an expanded convolutional layer, specifically by sequentially constructing a convolutional layer of size 3×3, a convolutional layer of size 6×6, and a convolutional layer of size 5×5, to construct the expanded convolutional layer;

[0114] Step S222: constructing a basic convolutional layer, specifically by constructing three convolutional layers of size 3×3 to construct the basic convolutional layer;

[0115] Step S23: constructing a classification model pooling layer, specifically, constructing a pooling layer after each dilated convolutional layer to perform an average pooling operation to construct the classification model pooling layer;

[0116] Step S24: constructing a fully connected layer of the classification model, specifically constructing three fully connected layers after the last convolution layer of the basic convolution layer to construct the fully connected layer of the classification model;

[0117] Step S25: constructing a classification loss function, specifically using a cross entropy loss function, to optimize the probability distribution;

[0118] Step S26: basic differentiation classification model training, specifically, through the differentiation category definition, the classification model deepening convolution layer, the classification model pooling layer, the classification model full connection layer and the classification loss function, the basic differentiation classification model training is performed based on the training sample set to obtain the basic differentiation classification model Model CNN ;

[0119] Step S27: basic differentiation classification, specifically using the basic differentiation classification model Model CNN , perform basic differentiation classification and obtain the basic probability image of cell differentiation.

[0120] By executing the above operations, in order to address the technical problem that in the existing stem cell differentiation prediction methods, the classification categories of the induced pluripotent stem cell differentiation process are insufficient, which affects the subsequent classification effect, this scheme creatively constructs nine differentiation categories and uses a convolutional neural network structure with a deepened convolutional layer for basic differentiation classification, realizing multi-category preliminary classification of cell differentiation classification, improving the versatility of the method, and optimizing the overall automation of the method.

[0121] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the cell region extraction is used to extract cell regions belonging to the same category. Specifically, a method based on a pre-trained densely connected convolutional network is used to extract cell regions based on the stem cell differentiation prediction original time-lapse microscopic image data and the stem cell differentiation category template image set to obtain cell region image data. The cell region image data is used to represent images of cell regions belonging to the same category but having a probability of being classified into different categories by the basic differentiation.

[0122] The pre-trained densely connected convolutional network includes a region extraction convolutional layer, a dense block, a transition layer, a region extraction pooling layer and a region extraction fully connected layer;

[0123] The region extraction convolution layer is used to extract image features;

[0124] The dense block is used to transmit feature map information;

[0125] The transition layer is used to reduce the dimension of the feature map;

[0126] The region extraction pooling layer is used to unify the size of feature map vectors;

[0127] The region extraction fully connected layer is used for classification and region extraction;

[0128] The method based on the pre-trained densely connected convolutional network is used to extract the cell region based on the original time-lapse microscopic image data of the stem cell differentiation prediction and the stem cell differentiation category template image set to obtain the cell region image data, including:

[0129] Step S31: constructing a region extraction convolution layer, specifically using a convolution kernel of size 1×1 to extract features, and using an S-type function for activation to construct the region extraction convolution layer;

[0130] Step S32: constructing a dense block, specifically constructing a three-layer dense block, using a ReLU activation function to activate each layer, and performing a convolution operation with a convolution kernel of size 3×3 at the end of each layer to construct the dense block;

[0131] Step S33: constructing a transition layer, specifically using a convolutional layer with a size of 1×1 and an average pooling layer with a size of 2×2 to perform a feature dimensionality reduction operation to construct the transition layer;

[0132] Step S34: construct a region extraction pooling layer, specifically, add a global pooling layer at the end of the dense block to convert the output feature map of the dense block into a vector of a fixed size of 512×512;

[0133] Step S35: constructing a region extraction fully connected layer, specifically, connecting the output of the region extraction pooling layer to the fully connected layer, performing feature image classification, and using a softmax classifier to perform a cell region extraction operation to construct the region extraction fully connected layer;

[0134] Step S36: region extraction model training, specifically, by constructing the region extraction convolution layer, the dense block, the transition layer, the region extraction pooling layer and the region extraction fully connected layer, and according to the stem cell differentiation category template image set, 600 training sample images are extracted from the original time-lapse microscopy image data of stem cell differentiation prediction by random cropping, and the region extraction model training is performed to obtain the region extraction model Model DN ;

[0135] Step S37: Cell region extraction, specifically using the region extraction model Model DN , perform cell region extraction and obtain cell region image data.

[0136] By performing the above operations, in order to address the technical problem that in existing stem cell differentiation prediction methods, classification based on single-frame images may classify cell regions belonging to the same category into different categories, this solution creatively adopts a method based on pre-trained densely connected convolutional networks to extract cell regions, extract cell regions of the same category in a unified manner, and optimize the overall accuracy of the method.

[0137] Example 5, see Figure 1 , Figure 2 , Figure 3 and Figure 5 This embodiment is based on the above embodiment. In step S4, the cell image segmentation is used to segment the cell image and extract cell features. Specifically, a U-shaped network is used to segment the original time-lapse microscopic image data of the stem cell differentiation prediction to obtain a segmented cell image.

[0138] The U-shaped network includes a downsampling subnet and an upsampling subnet;

[0139] The downsampling subnet is used to extract high-level abstract features;

[0140] The upsampling subnet is used to retain semantic information and perform feature upsampling and fusion;

[0141] The step of using a U-shaped network to perform cell image segmentation on the original time-lapse microscopic image data for stem cell differentiation prediction to obtain a segmented cell image comprises:

[0142] Step S41: constructing a downsampling subnet, specifically constructing four double-layer convolutional layers, including constructing a first downsampling convolutional layer, a second downsampling convolutional layer, a third downsampling convolutional layer and a fourth downsampling convolutional layer, setting a maximum pooling layer below each double-layer convolutional layer, and setting a connection layer below each maximum pooling layer, including a first connection layer, a second connection layer, a third connection layer and a fourth connection layer, the connection layer is used to connect with the convolutional layer in the upsampling subnet, perform a cell feature extraction operation, and construct the downsampling subnet;

[0143] Step S42: constructing an upsampling subnet, specifically constructing 5 double-layer convolution layers, including constructing a first upsampling convolution layer, a second upsampling convolution layer, a third upsampling convolution layer, a fourth upsampling convolution layer and a fifth upsampling convolution layer, and setting an upsampling layer below the first upsampling convolution layer, the second upsampling convolution layer, the third upsampling convolution layer and the fourth upsampling convolution layer, and setting a softmax classifier below the fifth upsampling convolution layer, the fifth upsampling convolution layer is connected to the maximum pooling layer of the first downsampling convolution layer through the first connection layer, the fourth upsampling convolution layer is connected to the maximum pooling layer of the second downsampling convolution layer through the second connection layer, the third upsampling convolution layer is connected to the maximum pooling layer of the third downsampling convolution layer through the third connection layer, and the second upsampling convolution layer is connected to the maximum pooling layer of the fourth downsampling convolution layer through the fourth connection layer, performing feature fusion operation, and constructing the upsampling subnet;

[0144] Step S43: constructing a cell segmentation model loss function, specifically comprising the following steps:

[0145] Step S431: construct a weight loss function, the calculation formula is:

[0146] ;

[0147] Where, L W is the weight loss function, N is the total number of samples, j is the sample index, and E is the cross entropy loss function. The calculation formula of the cross entropy loss function E is: E=-Y×logX, where Y is the true value of cell segmentation, X is the predicted value of cell segmentation, is the loss weight;

[0148] Step S432: construct a scatter loss function, the calculation formula is:

[0149] ;

[0150] Where, L D is the scatter loss function, X is the predicted value of cell segmentation, Y is the true value of cell segmentation, is the intersection operator, |·| is the modulus operator;

[0151] Step S433: construct a cell segmentation model loss function, the calculation formula is:

[0152] ;

[0153] Where L is the cell segmentation model loss function, is the weight loss balance coefficient, L W is the weight loss function, is the scatter loss balance coefficient, L Dis the scatter loss function;

[0154] Step S44: Cell segmentation model training, specifically, by constructing the downsampling subnet, constructing the upsampling subnet and constructing the cell segmentation model loss function, the cell segmentation model is trained to obtain the cell segmentation model Model UN ;

[0155] Step S45: Cell image segmentation, specifically using the cell segmentation model Model UN , perform cell image segmentation and obtain the segmented cell image.

[0156] By performing the above operations, in order to solve the technical problem that the existing stem cell differentiation prediction methods have low accuracy in cell image segmentation, which affects the subsequent calculation efficiency and accuracy, this solution creatively uses a U-shaped network for cell image segmentation, optimizes the recognition effect of specific types of cell areas, and provides a good data foundation for subsequent continuous time series prediction.

[0157] Example 6, see Figure 1 , Figure 2 and Figure 6 This embodiment is based on the above embodiment. In step S5, the stem cell differentiation prediction is used to predict stem cell differentiation, specifically using a spatiotemporal recurrent neural network to perform time-series stem cell differentiation prediction based on the cell differentiation basic probability image, the cell region image data and the segmented cell image to obtain continuous stem cell differentiation prediction data;

[0158] The spatiotemporal recurrent neural network comprises a feedforward training subnet, a pre-training subnet and a basic recurrent neural network;

[0159] The feedforward training subnet is used to predict the stem cell differentiation status of the next frame of the time series image at the current moment;

[0160] The pre-trained subnet is used to generate initial parameters of a basic recurrent neural network through training;

[0161] The basic recurrent neural network is used to predict stem cell differentiation;

[0162] The step of using a spatiotemporal recurrent neural network to perform time-series stem cell differentiation prediction based on the cell differentiation basic probability image, the cell region image data and the segmented cell image to obtain stem cell differentiation prediction data includes:

[0163] Step S51: constructing a feedforward training subnet, specifically by constructing a four-layer stacked long short-term memory neural network, combining the cell region image data and the segmented cell image to perform cell recognition, and performing a stem cell differentiation prediction operation for the next frame of the time series image based on the single frame image to obtain a sub-frame differentiation prediction image, and constructing the feedforward training subnet;

[0164] Step S52: constructing a pre-trained subnet, specifically using the continuous probability image in the basic probability image of cell differentiation to construct the pre-trained subnet and obtain the initial parameters of the basic recurrent neural network;

[0165] Step S53: constructing a basic recurrent neural network, specifically, based on the initial parameters of the basic recurrent neural network and the sub-frame differentiation prediction image, iteratively learning a four-layer stacked long short-term memory neural network to obtain a continuous stem cell differentiation prediction image, and constructing the basic recurrent neural network;

[0166] Step S54: continuous stem cell differentiation prediction model training, specifically, by constructing the feedforward training subnet, constructing the pre-training subnet and constructing the basic recurrent neural network, performing continuous stem cell differentiation prediction model training to obtain a continuous stem cell differentiation prediction model Model RNN ;

[0167] Step S55: Stem cell differentiation prediction, specifically using the continuous stem cell differentiation prediction model Model RNN Perform stem cell differentiation prediction to obtain continuous stem cell differentiation prediction data.

[0168] By performing the above operations, in order to address the technical problem that in the existing stem cell differentiation prediction methods, the stem cell differentiation prediction based on single-frame images cannot effectively reflect the complete process of cell differentiation and cannot meet the prediction needs, this solution creatively uses a spatiotemporal recurrent neural network to predict stem cell differentiation in continuous frames. By obtaining continuous stem cell differentiation prediction data, it better reflects the process of stem cell differentiation and the cell information in the reprogramming stage, thereby improving the overall usability of the method and broadening the scope of application of the method.

[0169] Embodiment 7, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, and the present invention provides a stem cell differentiation prediction system based on machine learning, including a data acquisition module, a basic differentiation classification module, a cell region extraction module, a cell image segmentation module and a stem cell differentiation prediction module;

[0170] The data acquisition module is used for data acquisition, and obtains original time-lapse microscopic image data for stem cell differentiation prediction through data acquisition, and sends the original time-lapse microscopic image data for stem cell differentiation prediction to the basic differentiation classification module, the cell region extraction module, the cell image segmentation module and the stem cell differentiation prediction module;

[0171] The basic differentiation classification module is used for basic differentiation classification, and obtains a stem cell differentiation category template image set and a cell differentiation basic probability image through basic differentiation classification, and sends the stem cell differentiation category template image set to the cell region extraction module, and sends the cell differentiation basic probability image to the stem cell differentiation prediction module;

[0172] The cell region extraction module is used for cell region extraction, obtains cell region image data through cell region extraction, and sends the cell region image data to the stem cell differentiation prediction module;

[0173] The cell image segmentation module is used for cell image segmentation, obtains a segmented cell image through cell image segmentation, and sends the segmented cell image to the stem cell differentiation prediction module;

[0174] The stem cell differentiation prediction module is used for stem cell differentiation prediction, and obtains continuous stem cell differentiation prediction data through stem cell differentiation prediction.

[0175] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0176] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0177] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for predicting stem cell differentiation based on machine learning, characterized in that: The method comprises the following steps: Step S1: data collection; Step S2: basic differentiation classification; Step S3: cell region extraction; Step S4: cell image segmentation; Step S5: stem cell differentiation prediction; In step S1, the data acquisition is used to acquire image data of induced pluripotent stem cells, specifically, using an incubator-type microscope to acquire original time-lapse microscopic image data for stem cell differentiation prediction through time-lapse acquisition; In step S2, the basic differentiation classification is used to define differentiation categories and generate probability images, specifically to construct nine differentiation categories, and use a convolutional neural network structure with a deepened convolutional layer to perform basic differentiation classification to obtain a basic probability image of cell differentiation; the convolutional neural network structure with a deepened convolutional layer includes a classification model deepened convolutional layer, a classification model pooling layer, and a classification model fully connected layer; In step S3, the cell region extraction is used to extract cell regions belonging to the same category, specifically, a method based on a pre-trained densely connected convolutional network is used to extract cell regions based on the original time-lapse microscopic image data for stem cell differentiation prediction and the stem cell differentiation category template image set to obtain cell region image data, wherein the cell region image data is used to represent images of cell regions belonging to the same category but having a probability of being basically differentiated and classified into different categories; the pre-trained densely connected convolutional network includes a region extraction convolutional layer, a dense block, a transition layer, a region extraction pooling layer and a region extraction fully connected layer; In step S4, the cell image segmentation is used to segment the cell image and extract cell features, specifically using a U-type network to perform cell image segmentation on the original time-lapse microscopic image data for stem cell differentiation prediction to obtain a segmented cell image; the U-type network includes a downsampling subnet and an upsampling subnet; In step S5, the stem cell differentiation prediction is used to perform stem cell differentiation prediction, specifically by using a spatiotemporal recurrent neural network to perform time-series stem cell differentiation prediction based on the cell differentiation basic probability image, the cell region image data and the segmented cell image to obtain continuous stem cell differentiation prediction data. The spatiotemporal recurrent neural network includes a feedforward training subnet, a pre-training subnet and a basic recurrent neural network.

2. The method for predicting stem cell differentiation based on machine learning according to claim 1, characterized in that: In step S2, the convolutional neural network structure of the deepened convolutional layer includes a classification model deepened convolutional layer, a classification model pooling layer and a classification model fully connected layer; The deepening convolution layer includes an expanded convolution layer and a basic convolution layer, wherein the expanded convolution layer is used to extract large-scale features, and the basic convolution layer is used to perform feature filtering on the extracted features; The classification model pooling layer is used to perform an average pooling operation; The fully connected layer of the classification model is used to calculate the probability of cell differentiation categories, specifically to calculate the classification probabilities of nine types of cell differentiation images.

3. The method for predicting stem cell differentiation based on machine learning according to claim 2, characterized in that: In step S2, the step of constructing nine differentiation categories and using a convolutional neural network structure with a deeper convolutional layer to perform basic differentiation classification to obtain a basic probability image of cell differentiation includes: Step S21: Differentiation category definition, specifically, nine categories of stem cell differentiation are defined for induced pluripotent stem cell images to obtain a stem cell differentiation category template image set I B , and according to the stem cell differentiation category template image set I B , by cutting out 2000 training sample sets from the original time-lapse microscopic image data for stem cell differentiation prediction; The stem cell differentiation category template image set I B , including background image I BG , Dispersed Single Cell Image I DSC , single cell images in colonies or non-colonies SC , floating cells in colonies or non-colonies I FC , intermediate cell image in the colony I CC , Image of cells in the early stages of differentiation of induced pluripotent stem cells I FPSC , Induced Pluripotent Stem Cell Image I PSC , Image of late differentiation of induced pluripotent stem cells I LPSC and inter-colony accumulation cell image I PBC , the calculation formula for the differentiation category definition is: I B ={I BG ,I DSC ,I SC ,I FC ,I CC ,I FPSC ,I PSC ,I LPSC ,I PBC }; In the formula, I B is a set of template images of stem cell differentiation category, I BG is a background image, I DSC is a dispersed single cell image, I SC is a single cell image in a colony or non-colony, I FC Is the image of floating cells in colonies or non-colonies, I CC is the middle cell image in the colony, I FPSC This is an image of cells in the early stages of differentiation of induced pluripotent stem cells. PSC This is an image of induced pluripotent stem cells. LPSC This is an image of the late stage of induced pluripotent stem cell differentiation. PBC It is an image of cells piled up between colonies; Step S22: Constructing a classification model to deepen the convolutional layer, specifically constructing 6 convolutional layers, of which 3 convolutional layers are used to construct dilated convolutional layers and 3 convolutional layers are used to construct basic convolutional layers, including the following steps: Step S221: constructing an expanded convolutional layer, specifically by sequentially constructing a convolutional layer of size 3×3, a convolutional layer of size 6×6, and a convolutional layer of size 5×5, to construct the expanded convolutional layer; Step S222: constructing a basic convolutional layer, specifically by constructing three convolutional layers of size 3×3 to construct the basic convolutional layer; Step S23: constructing a classification model pooling layer, specifically, constructing a pooling layer after each dilated convolutional layer to perform an average pooling operation to construct the classification model pooling layer; Step S24: constructing a fully connected layer of the classification model, specifically constructing three fully connected layers after the last convolution layer of the basic convolution layer to construct the fully connected layer of the classification model; Step S25: constructing a classification loss function, specifically using a cross entropy loss function, to optimize the probability distribution; Step S26: basic differentiation classification model training, specifically, through the differentiation category definition, the classification model deepening convolution layer, the classification model pooling layer, the classification model full connection layer and the classification loss function, the basic differentiation classification model training is performed based on the training sample set to obtain the basic differentiation classification model Model CNN ; Step S27: basic differentiation classification, specifically using the basic differentiation classification model Model CNN , perform basic differentiation classification and obtain the basic probability image of cell differentiation.

4. The method for predicting stem cell differentiation based on machine learning according to claim 3, characterized in that: In step S3, the pre-trained densely connected convolutional network includes a region extraction convolutional layer, a dense block, a transition layer, a region extraction pooling layer and a region extraction fully connected layer; The region extraction convolution layer is used to extract image features; The dense block is used to transmit feature map information; The transition layer is used to reduce the dimension of the feature map; The region extraction pooling layer is used to unify the size of feature map vectors; The region extraction fully connected layer is used for classification and region extraction; The method based on the pre-trained densely connected convolutional network is used to extract the cell region based on the original time-lapse microscopic image data of the stem cell differentiation prediction and the stem cell differentiation category template image set to obtain the cell region image data, including: Step S31: constructing a region extraction convolution layer, specifically using a convolution kernel of size 1×1 to extract features, and using an S-type function for activation to construct the region extraction convolution layer; Step S32: constructing a dense block, specifically constructing a three-layer dense block, using a ReLU activation function to activate each layer, and performing a convolution operation with a convolution kernel of size 3×3 at the end of each layer to construct the dense block; Step S33: constructing a transition layer, specifically using a convolutional layer with a size of 1×1 and an average pooling layer with a size of 2×2 to perform a feature dimensionality reduction operation to construct the transition layer; Step S34: construct a region extraction pooling layer, specifically, add a global pooling layer at the end of the dense block to convert the output feature map of the dense block into a vector of a fixed size of 512×512; Step S35: constructing a region extraction fully connected layer, specifically, connecting the output of the region extraction pooling layer to the fully connected layer, performing feature image classification, and using a softmax classifier to perform a cell region extraction operation to construct the region extraction fully connected layer; Step S36: region extraction model training, specifically, by constructing the region extraction convolution layer, the dense block, the transition layer, the region extraction pooling layer and the region extraction fully connected layer, and according to the stem cell differentiation category template image set, 600 training sample images are extracted from the original time-lapse microscopy image data of stem cell differentiation prediction by random cropping, and the region extraction model training is performed to obtain the region extraction model Model DN ; Step S37: Cell region extraction, specifically using the region extraction model Model DN , perform cell region extraction and obtain cell region image data.

5. The method for predicting stem cell differentiation based on machine learning according to claim 4, characterized in that: In step S4, the U-shaped network includes a downsampling subnet and an upsampling subnet; The downsampling subnet is used to extract high-level abstract features; The upsampling subnet is used to retain semantic information and perform feature upsampling and fusion; The step of using a U-shaped network to perform cell image segmentation on the original time-lapse microscopic image data for stem cell differentiation prediction to obtain a segmented cell image comprises: Step S41: constructing a downsampling subnet, specifically constructing four double-layer convolutional layers, including constructing a first downsampling convolutional layer, a second downsampling convolutional layer, a third downsampling convolutional layer and a fourth downsampling convolutional layer, setting a maximum pooling layer below each double-layer convolutional layer, and setting a connection layer below each maximum pooling layer, including a first connection layer, a second connection layer, a third connection layer and a fourth connection layer, the connection layer is used to connect with the convolutional layer in the upsampling subnet, perform a cell feature extraction operation, and construct the downsampling subnet; Step S42: constructing an upsampling subnet, specifically constructing 5 double-layer convolution layers, including constructing a first upsampling convolution layer, a second upsampling convolution layer, a third upsampling convolution layer, a fourth upsampling convolution layer and a fifth upsampling convolution layer, and setting an upsampling layer below the first upsampling convolution layer, the second upsampling convolution layer, the third upsampling convolution layer and the fourth upsampling convolution layer, and setting a softmax classifier below the fifth upsampling convolution layer, the fifth upsampling convolution layer is connected to the maximum pooling layer of the first downsampling convolution layer through the first connection layer, the fourth upsampling convolution layer is connected to the maximum pooling layer of the second downsampling convolution layer through the second connection layer, the third upsampling convolution layer is connected to the maximum pooling layer of the third downsampling convolution layer through the third connection layer, and the second upsampling convolution layer is connected to the maximum pooling layer of the fourth downsampling convolution layer through the fourth connection layer, performing feature fusion operation, and constructing the upsampling subnet; Step S43: constructing a cell segmentation model loss function, specifically comprising the following steps: Step S431: construct a weight loss function, the calculation formula is: Where, L W is a weighted loss function, N is the total number of samples, j is the sample index, and E is a cross entropy loss function, wherein the calculation formula of the cross entropy loss function E is: E=-Y×logX, where Y is the true value of cell segmentation, X is the predicted value of cell segmentation, and ω is the loss weight; Step S432: construct a scatter loss function, the calculation formula is: Where, L D is the scatter loss function, X is the predicted value of cell segmentation, Y is the true value of cell segmentation, ∩ is the intersection operator, and |·| is the modulus operator; Step S433: construct a cell segmentation model loss function, the calculation formula is: L=α×L W +β×L D ; Where L is the cell segmentation model loss function, α is the weight loss balance coefficient, and L W is the weight loss function, β is the scatter loss balance coefficient, L D is the scatter loss function; Step S44: Cell segmentation model training, specifically, by constructing the downsampling subnet, constructing the upsampling subnet and constructing the cell segmentation model loss function, the cell segmentation model is trained to obtain the cell segmentation model Model UN ; Step S45: Cell image segmentation, specifically using the cell segmentation model Model UN , perform cell image segmentation and obtain the segmented cell image.

6. The method for predicting stem cell differentiation based on machine learning according to claim 5, characterized in that: In step S5, the spatiotemporal recurrent neural network includes a feedforward training subnet, a pre-training subnet and a basic recurrent neural network; The feedforward training subnet is used to predict the stem cell differentiation status of the next frame of the time series image at the current moment; The pre-trained subnet is used to generate initial parameters of a basic recurrent neural network through training; The basic recurrent neural network is used to predict stem cell differentiation; The step of using a spatiotemporal recurrent neural network to perform time-series stem cell differentiation prediction based on the cell differentiation basic probability image, the cell region image data and the segmented cell image to obtain stem cell differentiation prediction data includes: Step S51: constructing a feedforward training subnet, specifically by constructing a four-layer stacked long short-term memory neural network, combining the cell region image data and the segmented cell image to perform cell recognition, and performing a stem cell differentiation prediction operation for the next frame of the time series image based on the single frame image to obtain a sub-frame differentiation prediction image, and constructing the feedforward training subnet; Step S52: constructing a pre-trained subnet, specifically using the continuous probability image in the basic probability image of cell differentiation to construct the pre-trained subnet and obtain the initial parameters of the basic recurrent neural network; Step S53: constructing a basic recurrent neural network, specifically, based on the initial parameters of the basic recurrent neural network and the sub-frame differentiation prediction image, iteratively learning a four-layer stacked long short-term memory neural network to obtain a continuous stem cell differentiation prediction image, and constructing the basic recurrent neural network; Step S54: continuous stem cell differentiation prediction model training, specifically, by constructing the feedforward training subnet, constructing the pre-training subnet and constructing the basic recurrent neural network, performing continuous stem cell differentiation prediction model training to obtain a continuous stem cell differentiation prediction model Model RNN ; Step S55: Stem cell differentiation prediction, specifically using the continuous stem cell differentiation prediction model Model RNN Perform stem cell differentiation prediction to obtain continuous stem cell differentiation prediction data.

7. The method for predicting stem cell differentiation based on machine learning according to claim 6, characterized in that: In step S1, the data acquisition is used to acquire induced pluripotent stem cell image data, specifically using an incubator-type microscope to obtain original time-lapse microscopic image data for predicting stem cell differentiation from the microscope imaging data through time-lapse acquisition. The original time-lapse microscopic image data for predicting stem cell differentiation specifically refers to frame-by-frame image data of stem cell differentiation time series.

8. A stem cell differentiation prediction system based on machine learning, used to implement a stem cell differentiation prediction method based on machine learning as claimed in any one of claims 1 to 7, characterized in that: It includes a data acquisition module, a basic differentiation classification module, a cell region extraction module, a cell image segmentation module and a stem cell differentiation prediction module.

9. The stem cell differentiation prediction system based on machine learning according to claim 8, characterized in that: The data acquisition module is used for data acquisition, and obtains original time-lapse microscopic image data for stem cell differentiation prediction through data acquisition, and sends the original time-lapse microscopic image data for stem cell differentiation prediction to the basic differentiation classification module, the cell region extraction module, the cell image segmentation module and the stem cell differentiation prediction module; The basic differentiation classification module is used for basic differentiation classification, and obtains a stem cell differentiation category template image set and a cell differentiation basic probability image through basic differentiation classification, and sends the stem cell differentiation category template image set to the cell region extraction module, and sends the cell differentiation basic probability image to the stem cell differentiation prediction module; The cell region extraction module is used for cell region extraction, obtains cell region image data through cell region extraction, and sends the cell region image data to the stem cell differentiation prediction module; The cell image segmentation module is used for cell image segmentation, obtains a segmented cell image through cell image segmentation, and sends the segmented cell image to the stem cell differentiation prediction module; The stem cell differentiation prediction module is used for stem cell differentiation prediction, and obtains continuous stem cell differentiation prediction data through stem cell differentiation prediction.

Images