Cell counting method and device and computer equipment

Abstract

The embodiment of the invention is suitable for the technical field of biomedicine and image processing, and provides a cell counting method and device and computer equipment, and the method comprises the steps: collecting an adherent stem cell image; segmenting the adherent stem cell image by using a detection segmentation model obtained by training to obtain a plurality of stem cell masks; determining the number of stem cells contained in the adherent stem cell image according to the plurality of stem cell masks; wherein the detection segmentation model comprises a backbone network structure, a convolution layer in the backbone network structure is a multi-layer pyramid convolution, and the multi-layer pyramid convolution is used for extracting multi-scale feature information in the adherent stem cell image. By adopting the method, the adherent stem cells can be rapidly and accurately counted.

Description

Technical Field

[0001] The embodiments of the present application belong to the fields of biomedical technology and image processing technology, and in particular, relate to a cell counting method, device and computer equipment. Background Art

[0002] Counting adherent stem cells is crucial for evaluating the growth status, proliferation ability and effectiveness of experimental operations of cells. By regularly counting adherent stem cells, the growth rate and growth status of cells can be monitored, which helps to optimize culture conditions and understand the biological characteristics of cells. Existing cell counting methods mainly include Coulter counting, manual counting and counting methods using flow cytometry, but there are some disadvantages in using these methods for cell counting. For example, the Coulter counting method is not applicable to stem cell research, and the use of manual counting or flow cytometry for cell counting is difficult to achieve when the amount of data is large. The counting results cannot be accurate to the individual digits, resulting in a low accuracy rate of the counting results.

[0003] Therefore, there is an urgent need for a method that can quickly and accurately count cells to overcome the related problems existing in the prior art. Summary of the invention

[0004] In view of this, the embodiments of the present application provide a cell counting method, apparatus and computer equipment for counting cells quickly and accurately.

[0005] A first aspect of an embodiment of the present application provides a cell counting method, comprising:

[0006] Collect images of adherent stem cells;

[0007] Segmenting the adherent stem cell image using the trained detection and segmentation model to obtain a plurality of stem cell masks;

[0008] determining the number of stem cells included in the adherent stem cell image according to the plurality of stem cell masks;

[0009] Among them, the detection and segmentation model includes a backbone network structure, and the convolution layer in the backbone network structure is a multi-layer pyramid convolution, and the multi-layer pyramid convolution is used to extract multi-scale feature information in the adherent stem cell image.

[0010] In the embodiment of the present application, before using the trained detection and segmentation model to segment the adherent stem cell image to obtain a plurality of stem cell masks, the method further includes:

[0011] Generate a data set for training the detection and segmentation model, wherein the data set includes a plurality of sample stem cell images annotated with mask labels, and the mask labels are used to indicate stem cell regions in the sample stem cell images;

[0012] Using the data set to train the pre-built model to obtain the detection and segmentation model;

[0013] Among them, the pre-built model includes the backbone network structure, the neck structure and the head structure. Each composite convolutional layer of the neck structure is configured with an attention mechanism layer. The neck structure is used to fuse and enhance the multi-scale feature information, and the head structure is used to output the stem cell mask in the sample stem cell image.

[0014] In a possible implementation manner of the embodiment of the present application, generating a data set for training the detection and segmentation model includes:

[0015] Training a pre-labeling model, wherein the pre-labeling model is obtained by predicting the stem cell mask of the sample stem cell image using a segmentation model to obtain a corresponding mask label, and then inputting the mask label of the sample stem cell image into the segmentation model for training;

[0016] Using the pre-labeling model to predict the plurality of sample stem cell images to obtain a mask label for each of the sample stem cell images;

[0017] The mask label of each sample stem cell image is matched with the corresponding sample stem cell image one by one to obtain the data set, and the data set is divided into a training data set, a verification data set and a test data set in proportion.

[0018] In an embodiment of the present application, before the pre-built model is trained using the data set to obtain the detection segmentation model, the method further includes:

[0019] The data set is subjected to data preprocessing, wherein the data preprocessing includes performing a data enhancement operation on the sample stem cell images in the data set, and using a heat diffusion algorithm to process the sample stem cell images after data enhancement.

[0020] In a possible implementation of the embodiment of the present application, the step of processing the sample stem cell image after data enhancement by using a heat diffusion algorithm includes:

[0021] Using the heat diffusion algorithm in each region of interest of each sample stem cell image annotated with a mask label, iteratively diffuses from the center of each region of interest to simulate and create a topological map of the stem cell region in each sample stem cell image;

[0022] The gradient vector fields in the x-direction and the y-direction of each sample stem cell image are obtained based on the topological map analysis, and the gradient vector fields are normalized.

[0023] The embodiment of the present application uses the heat diffusion algorithm in the process of image processing, and can use the data processed by the heat diffusion algorithm as input data for subsequent processing. In this way, the input label is processed by the heat diffusion algorithm to form a heat diffusion vector field, which is helpful for the subsequent use of the detection segmentation model to learn the image shape of the stem cell area with a star-shaped contour, and realize the accurate segmentation of the stem cell area in the stem cell image.

[0024] In the embodiment of the present application, the detection and segmentation model obtained by training is used to segment the adherent stem cell image to obtain a plurality of stem cell masks, including:

[0025] Extracting multi-scale feature information in the adherent stem cell image using multi-layer pyramid convolution in the detection and segmentation model;

[0026] Processing the multi-scale feature information using the neck structure of the detection and segmentation model to obtain an output feature map;

[0027] The head structure of the detection segmentation model is used to extract each stem cell mask in the output feature map.

[0028] In a possible implementation manner of the embodiment of the present application, the multi-scale feature information is processed using the neck structure of the detection segmentation model to obtain an output feature map, including:

[0029] Using the attention mechanism layer in the neck structure to perform a global average pooling operation on the multi-scale feature information;

[0030] Performing a one-dimensional convolution operation on the multi-scale feature information after the global average pooling operation, and determining the weight of each channel of the multi-scale feature information;

[0031] The weight is multiplied by the corresponding element in the original multi-scale feature information to obtain the output feature map.

[0032] In a possible implementation manner of the embodiment of the present application, after determining the number of stem cells included in the adherent stem cell image according to the plurality of stem cell masks, the method further includes:

[0033] Calculating the area of ​​each stem cell and the total area of ​​stem cells in the adherent stem cell image based on each of the stem cell masks;

[0034] The contour of each of the stem cells is determined respectively, and the circumference and circularity of each of the stem cells are calculated based on the contour.

[0035] A second aspect of the embodiments of the present application provides a cell counting device, comprising:

[0036] An acquisition module, used for acquiring images of adherent stem cells;

[0037] A segmentation module, used to segment the adherent stem cell image using the trained detection segmentation model to obtain a plurality of stem cell masks;

[0038] A determination module, configured to determine the number of stem cells contained in the adherent stem cell image according to the plurality of stem cell masks;

[0039] Among them, the detection and segmentation model includes a backbone network structure, and the convolution layer in the backbone network structure is a multi-layer pyramid convolution, and the multi-layer pyramid convolution is used to extract multi-scale feature information in the adherent stem cell image.

[0040] A third aspect of an embodiment of the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the computer device implements a method as described in any one of the first aspects above.

[0041] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a computer, the method described in any one of the first aspects above is implemented.

[0042] A fifth aspect of the embodiments of the present application provides a computer program product, including a computer program, which, when executed, enables the method described in any one of the first aspects to be executed.

[0043] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0044] In the embodiment of the present application, by improving the general or common detection network model, for example, improving the YO10-Seg model, the traditional convolution layer in the YO10-Seg model backbone network is replaced with a multi-layer pyramid convolution, so that when the collected adherent stem cell image is processed, the multi-layer pyramid convolution can be used to extract the multi-scale feature information in the adherent stem cell image. Through the data of the multi-scale feature information, the detection segmentation model can output an accurate stem cell mask, and then according to the multiple stem cell masks, the number of stem cells in the image can be determined, so as to achieve fast and accurate counting of adherent stem cells. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or prior art descriptions. Obviously, the drawings described below are only some embodiments of the present application, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0046] Figure 1 is a schematic diagram of a cell counting method provided in an embodiment of the present application;

[0047] Figure 2 is a schematic diagram of another cell counting method provided in an embodiment of the present application;

[0048] Figure 3 This is a schematic diagram of optimizing the image of adherent stem cells provided in an embodiment of the present application;

[0049] Figure 4 is a schematic diagram of a possible implementation of S201 in a cell counting method provided in an embodiment of the present application;

[0050] Figure 5 It is a schematic diagram of the structure of a pyramid convolution provided in an embodiment of the present application;

[0051] Figure 6 It is a structural diagram of an attention mechanism layer provided in an embodiment of the present application;

[0052] Figure 7 is a schematic diagram of a prediction result of a stem cell mask provided in an embodiment of the present application;

[0053] Figure 8 It is a schematic diagram of a process of cell counting and cell status analysis provided in an embodiment of the present application;

[0054] Fig. 9 is a schematic diagram of a cell counting device provided in an embodiment of the present application;

[0055] Fig.10 It is a schematic diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] In the following description, specific details such as specific system structures, technologies, etc. are proposed for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from hindering the description of the present application.

[0057] In response to various problems existing in existing cell counting methods, the industry has conducted research on possible counting methods based on various technical means.

[0058] For example, for stem cell counting, stem cells can be segmented based on the basic segmentation neural network FCN (Fully Convolutional Networks) in deep learning, and stem cell counting can be achieved based on segmentation. However, this technology belongs to the early semantic segmentation neural network, and the network performance is limited. It can only classify each pixel in the image as foreground or background, and cannot classify the foreground pixel to a cell of a certain instance. When faced with stem cells with adhesion, it is impossible to achieve accurate segmentation of stem cells.

[0059] For example, a related study provides a method for counting cells based on YOLOv3 and density estimation. This method can be used to detect the phenomenon that white blood cells are sparsely distributed and red blood cells are densely distributed in blood cell microscopic images. The YOLOv3 network model is first used to detect white blood cells, isolated red blood cells, and adhesion red blood cell areas in the microscopic image, and then the density estimation algorithm is used to count the red blood cells in the adhesion red blood cell area. This method supplements the density estimation algorithm of adhesion red blood cells on the basis of the Yolov3 detection algorithm. The main reason is that Yolov3 does not have the ability to accurately detect adhesion cells. However, the density estimation algorithm used in this method only estimates the number of adhesion cells and does not accurately count them. Neither Yolov3 nor the density estimation algorithm involves the segmentation of cell areas, and it is impossible to evaluate the cell state.

[0060] For example, in the algae cell statistics method based on deep learning and image pattern recognition, the deep learning detection network can be used to detect the location of the disc star algae cells and then use a series of traditional algorithms to determine the integrity of the disc star algae cells. However, since the traditional algorithm does not have generalization, multiple threshold parameters need to be adjusted during the application process. After changing the microscope shooting scene, this method may not be applicable. In addition, the process of applying this method, which requires first deep learning detection and then using traditional algorithm segmentation, is too cumbersome.

[0061] In response to various problems existing in the prior art and related research mentioned in the aforementioned examples, an embodiment of the present application provides a cell counting method, which improves common detection network models, such as the YOlO-Seg model, introduces pyramid convolution and attention mechanisms into the backbone network structure of the model, and improves the backbone network's ability to capture image features, thereby accurately segmenting the stem cell mask in the adherent stem cell image, and then being able to quickly and accurately count the adherent stem cells.

[0062] The technical solution of the present application is described below through specific embodiments.

[0063] Reference Figure 1 , shows a schematic diagram of a cell counting method provided in an embodiment of the present application, which may specifically include the following steps:

[0064] S101, collecting images of adherent stem cells.

[0065] It should be noted that the method can be implemented by a computer device, that is, the execution subject of the embodiment of the present application can be a computer device, and the computer device can achieve rapid and accurate counting of adherent stem cells by executing each step in the method. The computer device in the embodiment of the present application can be an electronic device with data processing capabilities, such as a desktop computer, a cloud server, etc. The embodiment of the present application does not limit the specific type of the electronic device.

[0066] The adherent stem cell images in the embodiments of the present application may be acquired at one or more time points during the cell culture process. The process of acquiring the adherent stem cell images may be achieved by a microscope or other equipment with similar functions.

[0067] Taking the acquisition of adherent stem cell images through a microscope as an example, in a possible implementation of the embodiment of the present application, the microscope can be integrated with the aforementioned computer device, that is, the microscope can be a part of the computer device, and after the adherent stem cell images are acquired, the data processing unit of the computer device processes them and outputs the corresponding counting results. Alternatively, the microscope can be independent of the aforementioned computer device, and after the adherent stem cell images are acquired through the microscope, the adherent stem cell images can be input into the computer device, processed by the computer device, and the corresponding counting results are output. The embodiment of the present application is not limited to this.

[0068] S102. Segment the adherent stem cell image using the trained detection and segmentation model to obtain a plurality of stem cell masks; the detection and segmentation model comprises a backbone network structure, wherein the convolutional layers in the backbone network structure are multi-layer pyramid convolutions, and the multi-layer pyramid convolutions are used to extract multi-scale feature information from the adherent stem cell image.

[0069] In the embodiment of the present application, the trained detection and segmentation model can be used to perform image segmentation on the adherent stem cells to obtain a plurality of stem cell masks, each of which represents a region where a stem cell is located.

[0070] The detection segmentation model in the embodiment of the present application can be improved and trained on the basis of a general or common detection network model. For example, the YO10-Seg model can be improved and the improved model can be trained to obtain a detection segmentation model that can be used for accurate image segmentation of adherent stem cell images.

[0071] Taking the YOlO-Seg model as an example, it mainly includes the backbone network structure, the neck structure and the head structure. Among them, the backbone network mainly realizes the feature extraction of the input image, including CBS (compound convolution), C2F (convolution to full connection) and SPPF (spatial pyramid pooling). CBS consists of a convolution layer (Conv2d), a batch normalization layer (BatchNorm2d) and an activation function.

[0072] One of the improvements of the above-mentioned YOlO-Seg model in the embodiment of the present application may include replacing all traditional convolutional layers in CBS with pyramid convolution (Pyconv). Pyramid convolution is a pyramid-like multi-layer structure proposed in recent years, which can be used to extract multi-scale feature information. Pyramid convolution includes a kernel pyramid, each layer includes different types of filters, and the size and depth of each filter are variable, so when using pyramid convolution to process an image, detail information of different scales can be extracted.

[0073] In an embodiment of the present application, the trained detection segmentation model can be obtained by replacing the traditional convolutional layer in the general YOlO-Seg model backbone network with a multi-layer pyramid convolution according to the above introduction, and then using the training data to train the improved model. In this way, when the detection segmentation model obtained by training is used to process the adherent stem cells, the replaced multi-layer pyramid convolution can be used to extract multi-scale feature information in the adherent stem cell image. On this basis, the neck structure in the model can further process the multi-scale feature information, and finally the head structure outputs multiple stem cell masks.

[0074] S103, determining the number of stem cells included in the adherent stem cell image according to the plurality of stem cell masks.

[0075] In the embodiment of the present application, each stem cell mask output by the detection segmentation model can represent a stem cell area. Therefore, based on the multiple stem cell masks output by the model, the number of stem cells contained in the currently acquired adherent stem cell image can be determined, thereby achieving rapid and accurate counting of adherent stem cells.

[0076] The embodiment of the present application improves the general or common detection network model, for example, improves the YO10-Seg model, replaces the traditional convolution layer in the YO10-Seg model backbone network with a multi-layer pyramid convolution, so that when the collected adherent stem cell image is processed, the multi-layer pyramid convolution can extract the multi-scale feature information in the adherent stem cell image. Through the data of the multi-scale feature information, the detection segmentation model can output an accurate stem cell mask, and then according to the multiple stem cell masks, the number of stem cells in the image can be determined, so as to achieve fast and accurate counting of adherent stem cells.

[0077] Reference Figure 2 , shows a schematic diagram of another cell counting method provided in an embodiment of the present application, which may specifically include the following steps:

[0078] S201 . Generate a data set for training a detection and segmentation model, wherein the data set includes a plurality of sample stem cell images annotated with mask labels.

[0079] In the embodiment of the present application, the data set used to train the detection segmentation model can be composed of multiple sample stem cell images marked with mask labels, and the sample stem cell images are images of adherent stem cells collected by a microscope or related equipment. That is, the images of adherent stem cells can be collected as samples for training the detection segmentation model.

[0080] In a possible implementation of the embodiment of the present application, the computer device may perform image optimization processing on the collected adherent stem cell image as a sample to enhance the pixel gradient change of the stem cell area image and highlight the stem cell image features.

[0081] Illustratively, the image optimization processing of the adherent stem cell image in the embodiment of the present application can mainly include processing the original image based on the opening operation and closing operation in the morphological image processing algorithm, and then optimizing the original image in combination with contrast enhancement, gamma correction and other brightness balancing means.

[0082] In image processing algorithms, the opening operation is a serial combination operation that first erodes the image and then dilates it; the closing operation is to first dilate the image and then erode it. The order of erosion and dilation in the opening and closing operations is opposite.

[0083] Among them, the processing process of the opening operation is to subtract the result of the image opening operation from the original image, so as to capture the peak information in the image. The opening operation is mainly aimed at bright objects on a dark background. It can correct the influence of uneven lighting, delete most of the non-uniform background, and separate the foreground and background of the image. The processing process of the closing operation is to subtract the result of the image closing operation from the original image, so as to capture the valley information in the image. The closing operation can highlight the dark details in the image and separate the background information that may exist in the image.

[0084] In the embodiment of the present application, in order to optimize the original image of adherent cells, the original image may be processed by an opening operation or a closing operation, or may be processed by an opening operation and a closing operation at the same time.

[0085] Exemplarily, the image of the adherent stem cells can be subjected to an opening operation and a closing operation in sequence to obtain a first image and a second image, respectively. That is, the first image is an image obtained after the computer device performs an opening operation on the image of the adherent stem cells, and the second image is an image obtained after the computer device performs a closing operation on the image of the adherent stem cells. Then, based on the original image of the adherent stem cells, an image obtained by subtracting the first image from the original image can be added, and an image obtained by subtracting the original image from the second image can be subtracted, thereby obtaining an optimized image of the adherent stem cells, thereby enhancing the information of the stem cell region in the original image.

[0086] The above process can be expressed as:

[0087] Target(I)=I+(I-opening(I))-(closing(I)-I)

[0088] Among them, Target(I) is the optimized image of adherent stem cells, I represents the original image of adherent stem cells, opening(I) represents the image obtained by opening the original image of adherent stem cells I, i.e., the first image, and closing(I) represents the image obtained by closing the original image of adherent stem cells I, i.e., the second image. Therefore, I-opening(I) represents the image obtained by subtracting the first image from the original image, and closing(I)-I represents the image obtained by subtracting the original image from the second image.

[0089] In a possible implementation of the embodiment of the present application, after the original image is processed based on the opening and closing operations in the morphological image processing algorithm to obtain an optimized wall-attached stem image, the computer device can perform subsequent processing on the wall-attached stem image to create a data set for training a detection and segmentation model.

[0090] In another possible implementation of the embodiment of the present application, after optimizing the original image of adherent stem cells by using opening and / or closing operations, the computer device may further optimize the image by processing means such as contrast enhancement and brightness balance, and use the further optimized image as the image for subsequent processing.

[0091] In the embodiment of the present application, when contrast enhancement processing is performed on an image, contrast enhancement algorithms such as histogram equalization and linear stretching can be used but not limited to be implemented. When brightness equalization processing is performed on an image, image brightness equalization adjustment can be implemented based on gamma correction. Among them, the gamma value (γ) is a nonlinear parameter that can be used to describe the relationship between input and output. By using a power function formula to describe the relationship between the input pixel value (Iin) and the output pixel value (Iout), brightness equalization processing of the image can be implemented. The above processing process can be expressed as:

[0092] Iout=Iin γ

[0093] like Figure 3 FIG. 1 is a schematic diagram of optimizing the image of adherent stem cells provided in an embodiment of the present application. Figure 3 (a) shows a schematic diagram of the original image of adherent stem cells. Figure 3 (b) shows the above-mentioned optimization methods for Figure 3 Schematic diagram of the image of adherent stem cells obtained after optimizing the original image shown in (a) in FIG. Figure 3 (b) shows the Figure 3 The effect diagram of (a) after optimization. Figure 3 As can be seen from (a) and (b) in the figure, in the image of adherent stem cells after optimization, the image features of stem cells are clearer and more prominent.

[0094] In the embodiment of the present application, the optimized adherent stem cell images can be used as samples to produce a data set that can be used to train a detection and segmentation model.

[0095] In a possible implementation of the embodiment of the present application, as Figure 4 As shown, generating a data set for training the detection and segmentation model in S201 may include the following steps S2011-S2013:

[0096] S2011, training a pre-labeling model, wherein the pre-labeling model is obtained by predicting the stem cell mask of the sample stem cell image using a segmentation model to obtain a corresponding mask label, and then inputting the mask label of the sample stem cell image into the segmentation model for training.

[0097] In an embodiment of the present application, the pre-labeling model may be a model for predicting the adherent stem cell images as samples to obtain the stem cell mask in each image. The pre-labeling model may be a model based on a general segmentation model, which is trained and fine-tuned to make the general segmentation model more suitable for the stem cell segmentation task in the adherent stem cell images.

[0098] In a possible implementation of the embodiment of the present application, the stem cell mask of the sample stem cell image can be predicted by using the segmentation model to obtain the corresponding mask label. The sample stem cell image is the sample adherent stem cell image, which can be the original adherent stem cell image according to Figure 3 The above-mentioned mask label can be used to indicate the stem cell area in the sample stem cell image.

[0099] After obtaining the mask label of the sample stem cell image, the erroneous stem cell contours in the prediction results can be adjusted to generate new mask labels.

[0100] The adjusted mask labels of the sample stem cell images can be input into the segmentation model for training, and the segmentation model can be fine-tuned to be a model that is more suitable for the stem cell segmentation task in the stem cell image, that is, the pre-labeled model in the embodiment of the present application.

[0101] S2012: Use the pre-labeling model to predict the multiple sample stem cell images to obtain a mask label for each sample stem cell image.

[0102] After the pre-labeling model is obtained, the pre-labeling model can be used to predict the collected adherent stem cell images as samples to obtain a mask label for each sample stem cell image.

[0103] S2013, making one-to-one correspondence between the mask label of each sample stem cell image and the corresponding sample stem cell image to obtain the data set, wherein the data set is divided into a training data set, a verification data set and a test data set in proportion.

[0104] By making a one-to-one correspondence between the mask label of each sample stem cell image and the corresponding sample stem cell image, a data set for training a detection and segmentation model can be produced.

[0105] In an embodiment of the present application, the produced data set can be divided into a training data set, a verification data set and a test data set according to a certain ratio, and used for training, verification and testing in the subsequent model training process.

[0106] For example, the prepared data set can be divided into a training data set, a validation data set, and a test data set in a ratio of 7:2:1. Therefore, the sample stem cell images used as the training data set account for 70% of the sample stem cell images in the entire data set, and the sample stem cell images used as the validation data set and the test data set account for 20% and 10% of the sample stem cell images in the entire data set, respectively.

[0107] In a possible implementation of the embodiment of the present application, before the pre-built model is trained using the data set to obtain the detection segmentation model, the data set may be pre-processed. The data pre-processing in the embodiment of the present application may include performing data enhancement operations on the sample stem cell images in the data set, and using a heat diffusion algorithm to process the sample stem cell images after data enhancement, and other processing processes.

[0108] For example, some data enhancement and augmentation operations may be performed on the sample stem cell images in the prepared data set. These operations may include but are not limited to image transformation processing such as brightness, contrast, cropping, and translation.

[0109] In an embodiment of the present application, for the sample stem cell image after data enhancement, a heat diffusion algorithm conversion can also be performed. The data processed by the heat diffusion algorithm can be used as input data for the next processing link. For example, the heat diffusion algorithm is used in each region of interest (ROI) of each sample stem cell image marked with a mask label, and iterative diffusion is performed from the center of each region of interest to simulate the creation of a topological map of the stem cell area in each sample stem cell image; based on the topological map, the gradient vector field in the x-direction and y-direction of each sample stem cell image can be parsed, and finally the gradient vector field is normalized. The image data obtained after normalization can be used for the segmentation branch to learn the image shape of stem cells with a star-shaped outline, improve the ability of the segmentation branch to accurately segment the stem cell area in the stem cell image, and achieve accurate segmentation of the stem cell area.

[0110] The normalization process may include but is not limited to minimum-maximum normalization, Z-score normalization, mean normalization, logarithmic normalization, positive and negative tangent normalization, etc. Taking minimum-maximum normalization as an example, the following formula shows an example of normalization of the gradient vector field in the x direction:

[0111]

[0112] Wherein, x′ is the normalized gradient vector field in the x direction, x is the gradient vector field in the x direction before normalization, max(x) and min(x) are the maximum and minimum values ​​of the gradient vector field in the x direction, respectively.

[0113] Therefore, the data set used to train the detection and segmentation model may be a sample stem cell image obtained by processing the image according to the aforementioned embodiments.

[0114] S202. Use the data set to train the pre-built model to obtain the detection segmentation model; wherein the pre-built model includes a backbone network structure, a neck structure and a head structure, the convolution layer in the backbone network structure is a multi-layer pyramid convolution, and each composite convolution layer of the neck structure is configured with an attention mechanism layer.

[0115] In an embodiment of the present application, a model can be pre-built and trained using the data set obtained after the above-mentioned processing, so as to obtain a detection segmentation model that can be used to segment the adherent stem cell image. The data set obtained after the above-mentioned processing can be the data output after being processed by the heat diffusion algorithm. The above-mentioned pre-built model may include a backbone network structure, a neck structure and a head structure. Among them, the convolution layer in the backbone network structure can be a multi-layer pyramid convolution, which is used to extract multi-scale feature information in the adherent stem cell image; the neck structure can be used to fuse and enhance multi-scale feature information; the head structure can be used to output the stem cell mask in the sample stem cell image. In the neck structure of the pre-built model in the embodiment of the present application, an attention mechanism layer can be configured after each composite convolution layer CBS. In this way, it can help the model to better focus on key information and ignore irrelevant information, thereby improving detection accuracy.

[0116] In a possible implementation of an embodiment of the present application, the pre-built model may be obtained by improving a general or common detection module, for example, by improving the YOlO-Seg model to obtain a model to be trained.

[0117] The YOlO-Seg model mainly includes the backbone network structure, the neck structure and the head structure. Among them:

[0118] The backbone network mainly realizes the feature extraction of the input image, including modules such as CBS, C2F and SPPF. CBS consists of a convolutional layer (Conv2d), a batch normalization layer (BatchNorm2d) and an activation function (for example, SiLU function). Conv2d is responsible for extracting features from the image, and gradually reduces the size of the feature map through downsampling operations while increasing the number of channels. C2F is a module for deep feature extraction. The SPPF module effectively reduces the amount of calculation by adopting a fast spatial pyramid pooling strategy. The SPPF module can enable the model to better adapt to scale changes by fusing multi-scale feature information, thereby improving the robustness of detection. The embodiment of the present application replaces all traditional convolutional layers in CBS with pyramid convolutions, which can extract multi-scale feature information. The pyramid convolution is a multi-layer structure, including a kernel pyramid. Each layer contains different types of filters. The size and depth of the filter are variable, so it can be used to extract detail information of different scales. The structure of the pyramid convolution can be as follows Figure 5 As shown in Figure 2, each layer of the multi-layer pyramid convolution can process the input feature map and output multi-scale feature information respectively.

[0119] The neck structure is located between the backbone network structure and the head structure, and its main function is to further fuse and enhance features. The neck structure of the YOlO-Seg model adopts a hybrid structure of FPN (Feature Pyramid Networks) and PAN (Path Aggregation Network) to achieve the fusion and transmission of multi-scale features. In the embodiment of the present application, a new attention mechanism layer is added after each CBS layer in the neck structure. The newly added attention mechanism can help the model better focus on key information and ignore irrelevant information, thereby improving detection accuracy. Exemplarily, the attention mechanism layer in the embodiment of the present application can be an effective channel attention (Efficient Channel Attention Module, ECA), and the structure of the ECA attention mechanism can be as follows Figure 6 shown.

[0120] In a possible implementation of the embodiment of the present application, combined with Figure 6 As shown, the process of applying the above attention mechanism can include the following steps:

[0121] ①Perform global average pooling operation on the input feature map;

[0122] ② Perform a one-dimensional convolution operation with a convolution kernel size of k, and obtain the weights of each channel through the Sigmoid activation function; for example, Figure 6 k = 5;

[0123] ③Multiply the weights by the corresponding elements of the original input feature map to obtain the final output feature map.

[0124] The head structure is the output part of the YOlO-Seg model, which is responsible for generating the bounding box of the target detection based on the extracted feature map, and performing processing such as category prediction and mask segmentation.

[0125] For the model constructed in the above manner, the computer device may use the prepared data set to perform model training. The steps of model training may include:

[0126] ① Forward propagation: The input data is calculated through the model to obtain the predicted output.

[0127] ② Calculate loss: Use the loss function to calculate the difference between the predicted output and the true label.

[0128] ③ Back propagation: Calculate the gradient based on the loss value and propagate it back to the network to update the weights.

[0129] ④Weight update: Use the optimizer to update the model’s weights based on the gradient.

[0130] After the pre-built model has completed the relevant training process, it can be used as a detection and segmentation model. In the subsequent image processing process, mask segmentation is performed on the adherent stem cell image to obtain multiple stem cell masks in each adherent stem cell image. Each stem cell mask can represent a specific stem cell in the image.

[0131] S203, collecting images of adherent stem cells.

[0132] In the embodiment of the present application, the adherent stem cell image is the image of stem cells to be counted or to be subjected to cell status analysis, and the adherent stem cell image can be acquired through a microscope or other equipment.

[0133] S204, extracting multi-scale feature information in the adherent stem cell image using the multi-layer pyramid convolution in the detection and segmentation model.

[0134] The detection segmentation model in the embodiment of the present application is a model obtained by training according to the above S201-S202, and the convolution layer in the backbone network of the detection segmentation model is a multi-layer pyramid convolution. The adherent stem cell image collected in S203 can be input into the above detection segmentation model for processing. Among them, the multi-layer pyramid convolution in the backbone network can be used to extract multi-scale feature information in the input adherent stem cell image.

[0135] S205 , using the neck structure of the detection and segmentation model to process the multi-scale feature information to obtain an output feature map.

[0136] In the embodiment of the present application, the neck structure in the detection and segmentation model can be further fused and enhanced for the extracted multi-scale feature information to achieve the fusion and transmission of multi-scale features.

[0137] In a possible implementation of an embodiment of the present application, an attention mechanism layer can be added behind each CBS layer in the neck structure to focus on key information in the features and ignore irrelevant information, thereby improving detection accuracy. Specifically, the attention mechanism layer in the neck structure can be used to perform a global average pooling operation on the multi-scale feature information, and then a one-dimensional convolution operation is performed on the multi-scale feature information after the global average pooling operation, and the weights of each channel of the multi-scale feature information are determined. By multiplying the weights of each channel with the corresponding elements in the original multi-scale feature information, the final output feature map can be obtained. The output feature map is the feature map of each stem cell mask that will be output after being processed by the head structure.

[0138] S206: Using the head structure of the detection and segmentation model to extract each stem cell mask in the output feature map.

[0139] In the embodiment of the present application, the head structure in the detection and segmentation model is responsible for generating a bounding box for target detection, neck category prediction and mask segmentation according to the extracted feature map. Therefore, for the feature map output by S205, the corresponding stem cell mask can be output by processing the head structure, and each stem cell mask represents a stem cell region in the original adherent stem cell image.

[0140] The process described in S204-S206 is to use the trained detection and segmentation model to predict the collected adherent stem cell images and output a single stem cell region mask. Figure 7 , which is a schematic diagram of the prediction result of a stem cell mask provided in an embodiment of the present application. Figure 7 (a) shows an image of adherent stem cells. Figure 7 (b) shows the use of the aforementioned detection segmentation model to Figure 7 (a) is a schematic diagram of the mask result output after prediction of the adherent stem cell image.

[0141] S207 , determining the number of stem cells included in the adherent stem cell image according to the plurality of stem cell masks.

[0142] In the embodiment of the present application, the result output by the detection segmentation model is n stem cell masks, each of which corresponds to a specific stem cell region in the image. Therefore, the number of stem cells contained in the original acquired adherent stem cell image can be determined based on the number of stem cell masks output by the model.

[0143] In a possible implementation of the embodiment of the present application, based on the stem cell mask output by the detection segmentation model, the computer device can also analyze the state of the stem cells. That is, in addition to determining the number of stem cells, the confluence of stem cells, the area, perimeter, roundness and other parameters of a single stem cell can also be output for analyzing the growth state of the stem cells.

[0144] In the embodiment of the present application, the area of ​​each stem cell and the total area of ​​stem cells in the adherent stem cell image can be calculated based on each stem cell mask, and by determining the contour of each stem cell respectively, the circumference and roundness of each stem cell and other parameters can be calculated based on the contour.

[0145] For example, each stem cell output by the detection and segmentation model can be traversed, and the number of pixels with a value of 1 (or any value representing the foreground) can be counted as the calculated area of ​​a single stem cell; the total area of ​​stem cells in the stem cell image, that is, the confluence of stem cells, can be obtained by accumulating the areas of single stem cells; in addition, the contour detection algorithm of image processing can be used to obtain the contour of a single stem cell, and the circumference of the stem cell can be calculated based on the contour of the stem cell; the roundness parameter of the stem cell can be further calculated based on the circumference of the stem cell, and the roundness is an indicator that describes the degree to which the shape of an object is close to a circle. The roundness calculation formula can be expressed as follows:

[0146]

[0147] The embodiment of the present application is aimed at the characteristics of stem cell images, and can be improved based on the common detection network model YOlO-Seg. The pyramid convolution and attention mechanism are introduced in the backbone network of the model, which effectively improves the ability of the backbone network to capture image features. Secondly, by using the heat diffusion algorithm in the segmentation branch of the model, the input label is processed to form a heat diffusion vector field, which is more helpful for the segmentation branch to learn the star-shaped contour of the stem cell area, and then the detection segmentation network is used to realize the accurate segmentation of the stem cell area in the adherent stem cell image, which effectively improves the model segmentation performance. Third, the embodiment of the present application uses the closed operation and open operation in the morphological image processing algorithm to process the image, and then combines the brightness balance means such as contrast enhancement and gamma correction to optimize the original image, strengthen the pixel gradient change of the stem cell area image, highlight the characteristics of the stem cell image, and use the segmentation model to fine-tune the training pre-labeling model, which also reduces the workload of labeling. Fourth, after obtaining the stem cell contour through the detection and segmentation model, the stem cell contour can be analyzed, and the contour parameters such as area, roundness, and circumference can be calculated through image processing technology, which enriches the types of parameters that can be output by the model and can assist users such as cell culture engineers to make more accurate judgments on the cell status.

[0148] For ease of understanding, the following is a complete example to introduce the cell counting and cell state analysis process provided in the embodiment of the present application. Figure 8 FIG. 1 is a flow chart of a cell counting and cell status analysis method provided in an embodiment of the present application. Figure 8 The process shown mainly includes:

[0149] Acquiring images of adherent stem cells: Images of adherent stem cells to be identified can be acquired by microscopy at one or more time points during the cell culture process.

[0150] Stem cell image optimization: For the adherent stem cell images collected in the above steps, the original images can be processed based on the opening and closing operations in the morphological image processing algorithm, and then combined with contrast enhancement, gamma correction and other brightness balancing methods to optimize the original images, strengthen the pixel gradient changes of the stem cell area image, and highlight the stem cell image characteristics.

[0151] Training the pre-labeling model: Based on the commonly used general segmentation model, the stem cell area mask in the stem cell image can be predicted, and the erroneous stem cell contours in the prediction results can be adjusted to generate new mask labels. The mask labels are then input into the general segmentation model for training, and the general segmentation model can be fine-tuned into a pre-labeling model that is more suitable for the stem cell segmentation task in stem cell images.

[0152] Dataset creation: The pre-labeled model trained in the previous step can be used to predict all the collected stem cell images, obtain mask labels, and create a network training dataset by matching the mask labels with the original images one by one. The dataset can be divided into a training dataset, a validation dataset, and a test dataset in a ratio of 7:2:1.

[0153] Dataset preprocessing: Before sending the dataset to the detection and segmentation model, the data can be enhanced and augmented, including but not limited to image transformations such as brightness, contrast, cropping, and translation. The enhanced data can be converted using the heat diffusion algorithm. In the annotated mask label, the heat diffusion algorithm is used for each ROI, and iterative diffusion is performed from the center of the ROI to simulate and create a topological map, parse the gradient vector fields in the x and y directions, and normalize the gradient vector fields.

[0154] Detection and segmentation model construction: It can be improved based on the classic detection neural network YOlO-Seg. All traditional convolutional layers in the CBS of the backbone network can be replaced with multi-layer pyramid convolutions to extract multi-scale feature information in the image; and a feature attention mechanism layer is added after each CBS layer of the neck structure to help the model better focus on key information and ignore irrelevant information, thereby improving detection accuracy.

[0155] Detection and segmentation model training: The detection and segmentation model constructed can be trained using the dataset prepared above.

[0156] Mask prediction and calculation of related parameters: Use the trained detection and segmentation model to predict the stem cell area in the stem cell image, obtain a single stem cell area mask, and then obtain the number of stem cells in the image. On this basis, traverse each mask, and use the number of pixels with a statistical value of 1 (or any value representing the foreground) as the calculated single stem cell area. Adding up the single stem cell area can obtain the total area of ​​all stem cells in the stem cell image, i.e., the stem cell confluence; use the image processing contour detection algorithm to obtain the single stem cell contour, and the stem cell circumference and roundness can be calculated based on the stem cell contour. The above-mentioned stem cell image may refer to an image obtained by collecting images of stem cells that require cell counting or cell status analysis.

[0157] Output related parameters: The detection and segmentation model can output parameters such as the number of stem cells, confluence, area, circumference, and roundness of a single stem cell in the stem cell image for relevant personnel to analyze the growth status of stem cells.

[0158] Reference Fig. 9 , shows a schematic diagram of a cell counting device provided in an embodiment of the present application, which may specifically include a collection module 901, a segmentation module 902 and a determination module 903, wherein:

[0159] An acquisition module 901 is used to acquire images of adherent stem cells;

[0160] A segmentation module 902 is used to segment the adherent stem cell image using the trained detection segmentation model to obtain a plurality of stem cell masks;

[0161] A determination module 903 is used to determine the number of stem cells included in the adherent stem cell image according to the plurality of stem cell masks;

[0162] Among them, the detection and segmentation model includes a backbone network structure, and the convolution layer in the backbone network structure is a multi-layer pyramid convolution, and the multi-layer pyramid convolution is used to extract multi-scale feature information in the adherent stem cell image.

[0163] In the embodiment of the present application, the device further includes a data set generation module and a detection and segmentation model training module.

[0164] A data set generation module, used to generate a data set for training the detection and segmentation model, wherein the data set includes a plurality of sample stem cell images annotated with mask labels, wherein the mask labels are used to indicate stem cell regions in the sample stem cell images;

[0165] A detection and segmentation model training module, used to train the pre-built model using the data set to obtain the detection and segmentation model;

[0166] Among them, the pre-built model includes the backbone network structure, the neck structure and the head structure. Each composite convolutional layer of the neck structure is configured with an attention mechanism layer. The neck structure is used to fuse and enhance the multi-scale feature information, and the head structure is used to output the stem cell mask in the sample stem cell image.

[0167] In a possible implementation of the embodiment of the present application, the data set generation module may be specifically used to:

[0168] Training a pre-labeling model, wherein the pre-labeling model is obtained by predicting the stem cell mask of the sample stem cell image using a segmentation model to obtain a corresponding mask label, and then inputting the mask label of the sample stem cell image into the segmentation model for training;

[0169] Using the pre-labeling model to predict the plurality of sample stem cell images to obtain a mask label for each of the sample stem cell images;

[0170] The mask label of each sample stem cell image is matched with the corresponding sample stem cell image one by one to obtain the data set, and the data set is divided into a training data set, a verification data set and a test data set in proportion.

[0171] In another possible implementation of the embodiment of the present application, the data set generation module may also be used to:

[0172] The data set is subjected to data preprocessing, wherein the data preprocessing includes performing a data enhancement operation on the sample stem cell images in the data set, and using a heat diffusion algorithm to process the sample stem cell images after data enhancement.

[0173] In the embodiment of the present application, the data set generation module may also be used for:

[0174] Using the heat diffusion algorithm in each region of interest of each sample stem cell image annotated with a mask label, iteratively diffuses from the center of each region of interest to simulate and create a topological map of the stem cell region in each sample stem cell image;

[0175] The gradient vector fields in the x-direction and the y-direction of each sample stem cell image are obtained based on the topological map analysis, and the gradient vector fields are normalized.

[0176] In the embodiment of the present application, the segmentation module 902 may be specifically used for:

[0177] Extracting multi-scale feature information in the adherent stem cell image using multi-layer pyramid convolution in the detection and segmentation model;

[0178] Processing the multi-scale feature information using the neck structure of the detection and segmentation model to obtain an output feature map;

[0179] The head structure of the detection segmentation model is used to extract each stem cell mask in the output feature map.

[0180] In a possible implementation of the embodiment of the present application, the segmentation module 902 may also be used to:

[0181] Using the attention mechanism layer in the neck structure to perform a global average pooling operation on the multi-scale feature information;

[0182] Performing a one-dimensional convolution operation on the multi-scale feature information after the global average pooling operation, and determining the weight of each channel of the multi-scale feature information;

[0183] The weight is multiplied by the corresponding element in the original multi-scale feature information to obtain the output feature map.

[0184] In the embodiment of the present application, the determining module 903 may also be used to:

[0185] Calculating the area of ​​each stem cell and the total area of ​​stem cells in the adherent stem cell image based on each of the stem cell masks;

[0186] The contour of each of the stem cells is determined respectively, and the circumference and circularity of each of the stem cells are calculated based on the contour.

[0187] An embodiment of the present application provides a cell counting device, and by using the device, each step in the aforementioned method embodiments can be implemented.

[0188] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment part.

[0189] Reference Fig.10 , shows a schematic diagram of a computer device provided by an embodiment of the present application. Fig.10 As shown, the computer device 1000 in the embodiment of the present application includes: a processor 1010, a memory 1020, and a computer program 1021 stored in the memory 1020 and executable on the processor 1010. When the processor 1010 executes the computer program 1021, the steps in each embodiment of the above-mentioned cell counting method are implemented, for example Figure 1Alternatively, when the processor 1010 executes the computer program 1021, the functions of each module / unit in the above-mentioned device embodiments are realized, for example Fig. 9 Functions of modules 901 to 903 are shown.

[0190] Exemplarily, the computer program 1021 may be divided into one or more modules / units, which are stored in the memory 1020 and executed by the processor 1010 to complete the present application. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which may be used to describe the execution process of the computer program 1021 in the computer device 1000. For example, the computer program 1021 may be divided into a collection module, a segmentation module, and a determination module, and the specific functions of each module are as follows:

[0191] An acquisition module, used for acquiring images of adherent stem cells;

[0192] A segmentation module, used to segment the adherent stem cell image using the trained detection segmentation model to obtain a plurality of stem cell masks;

[0193] A determination module, configured to determine the number of stem cells contained in the adherent stem cell image according to the plurality of stem cell masks;

[0194] Among them, the detection and segmentation model includes a backbone network structure, and the convolution layer in the backbone network structure is a multi-layer pyramid convolution, and the multi-layer pyramid convolution is used to extract multi-scale feature information in the adherent stem cell image.

[0195] The computer device 1000 may be a device capable of implementing the steps in the above-mentioned embodiments, and may be a desktop computer, a cloud server, or other devices. The computer device 1000 may include, but is not limited to, a processor 1010 and a memory 1020. Those skilled in the art may understand that Fig.10 This is only an example of the computer device 1000 and does not constitute a limitation of the computer device 1000. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device 1000 may also include input and output devices, network access devices, buses, etc.

[0196] The processor 1010 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0197] The memory 1020 may be an internal storage unit of the computer device 1000, such as a hard disk or memory of the computer device 1000. The memory 1020 may also be an external storage device of the computer device 1000, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device 1000. Further, the memory 1020 may also include both an internal storage unit of the computer device 1000 and an external storage device. The memory 1020 is used to store the computer program 1021 and other programs and data required by the computer device 1000. The memory 1020 may also be used to temporarily store data that has been output or is to be output.

[0198] An embodiment of the present application further discloses a computer 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, the methods described in the above embodiments are implemented.

[0199] The embodiments of the present application further disclose a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a computer, the methods described in the above embodiments are implemented.

[0200] The embodiments of the present application further disclose a computer program product, including a computer program. When the computer program is run on a computer, the computer is enabled to execute the methods described in the aforementioned embodiments.

[0201] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application is described in detail with reference to the above-mentioned embodiments, a person skilled in the art should understand that the technical solutions described in the above-mentioned embodiments can still be modified, or some of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A cell counting method, characterized in that: include: Collect images of adherent stem cells; Segmenting the adherent stem cell image using the trained detection and segmentation model to obtain a plurality of stem cell masks; determining the number of stem cells included in the adherent stem cell image according to the plurality of stem cell masks; Among them, the detection and segmentation model includes a backbone network structure, and the convolution layer in the backbone network structure is a multi-layer pyramid convolution, and the multi-layer pyramid convolution is used to extract multi-scale feature information in the adherent stem cell image.

2. The method according to claim 1, characterized in that Before segmenting the adherent stem cell image using the trained detection segmentation model to obtain a plurality of stem cell masks, the method further includes: Generate a data set for training the detection and segmentation model, wherein the data set includes a plurality of sample stem cell images annotated with mask labels, and the mask labels are used to indicate stem cell regions in the sample stem cell images; The pre-built model is trained using the data set to obtain the detection and segmentation model; Among them, the pre-built model includes the backbone network structure, the neck structure and the head structure. Each composite convolutional layer of the neck structure is configured with an attention mechanism layer. The neck structure is used to fuse and enhance the multi-scale feature information, and the head structure is used to output the stem cell mask in the sample stem cell image.

3. The method according to claim 2, characterized in that The generating of a data set for training the detection segmentation model comprises: Training a pre-labeling model, wherein the pre-labeling model is obtained by predicting the stem cell mask of the sample stem cell image using a segmentation model to obtain a corresponding mask label, and then inputting the mask label of the sample stem cell image into the segmentation model for training; Using the pre-labeling model to predict the plurality of sample stem cell images, to obtain a mask label for each of the sample stem cell images; The mask label of each sample stem cell image is matched with the corresponding sample stem cell image one by one to obtain the data set, and the data set is divided into a training data set, a verification data set and a test data set in proportion.

4. The method according to claim 2 or 3, characterized in that: Before using the data set to train the pre-built model to obtain the detection segmentation model, the method further includes: The data set is subjected to data preprocessing, wherein the data preprocessing includes performing a data enhancement operation on the sample stem cell images in the data set, and using a heat diffusion algorithm to process the sample stem cell images after data enhancement.

5. The method according to claim 4, characterized in that The step of processing the sample stem cell image after data enhancement by using a heat diffusion algorithm comprises: Using the heat diffusion algorithm in each region of interest of each sample stem cell image annotated with a mask label, iteratively diffuses from the center of each region of interest to simulate and create a topological map of the stem cell region in each sample stem cell image; The gradient vector fields in the x-direction and the y-direction of each sample stem cell image are obtained based on the topological map analysis, and the gradient vector fields are normalized.

6. The method according to any one of claims 1 to 3 or 5, characterized in that: The method of segmenting the adherent stem cell image using the trained detection segmentation model to obtain a plurality of stem cell masks includes: Extracting multi-scale feature information in the adherent stem cell image using multi-layer pyramid convolution in the detection and segmentation model; Processing the multi-scale feature information using the neck structure of the detection and segmentation model to obtain an output feature map; The head structure of the detection segmentation model is used to extract each stem cell mask in the output feature map.

7. The method according to claim 6, characterized in that The multi-scale feature information is processed using the neck structure of the detection segmentation model to obtain an output feature map, including: Using the attention mechanism layer in the neck structure to perform a global average pooling operation on the multi-scale feature information; Performing a one-dimensional convolution operation on the multi-scale feature information after the global average pooling operation, and determining the weight of each channel of the multi-scale feature information; The weight is multiplied by the corresponding element in the original multi-scale feature information to obtain the output feature map.

8. The method according to any one of claims 1 to 3 or 5 or 7, characterized in that: After determining the number of stem cells included in the adherent stem cell image according to the plurality of stem cell masks, the method further includes: Calculating the area of ​​each stem cell and the total area of ​​stem cells in the adherent stem cell image based on each of the stem cell masks; The contour of each of the stem cells is determined respectively, and the circumference and circularity of each of the stem cells are calculated based on the contour.

9. A computer 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, the computer device is caused to implement the method according to any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed, the method according to any one of claims 1 to 8 is executed.

Images