Image recognition method and related model training method and related devices and equipment

Abstract

The present application discloses an image recognition method and a training method of a related model and related devices and equipment, wherein the training method of a cell detection model includes: using a cell detection model to perform target detection on a sample image to obtain a detection result, wherein the detection result includes a detection area of ​​a target cell in the sample image, and the sample image is annotated with an actual area where the target cell is located; based on the difference between the detection area and the actual area and the suspected area of ​​the target cell, respectively, the loss value of the cell detection model is determined; using the loss value, the parameters of the cell detection model are adjusted to complete the current training of the cell detection model. The above scheme improves the accuracy of cell detection.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an image recognition method and a training method for related models and related devices and equipment. Background Art

[0002] With the development of artificial intelligence technologies such as neural networks and deep learning, training neural network models and using the trained neural network models to meet relevant business needs in the medical field have gradually become popular.

[0003] Among the relevant business needs, the detection of target cells such as signet ring cells has attracted attention due to its application value in improving patient survival rates. However, manual detection using a microscope is prone to omissions and misdetection, making it difficult to ensure the accuracy of target cell detection. In view of this, how to improve the accuracy of cell detection has become an urgent problem to be solved. Summary of the invention

[0004] The present application provides an image recognition method and a related model training method and related devices and equipment.

[0005] The first aspect of the present application provides a training method for a cell detection model, comprising: using the cell detection model to perform target detection on a sample image to obtain a detection result, wherein the detection result includes a detection area of ​​the target cell in the sample image, and the sample image is annotated with an actual area where the target cell is located; based on the differences between the detection area and the actual area and the suspected area of ​​the target cell, respectively, determining a loss value of the cell detection model; using the loss value, adjusting the parameters of the cell detection model to complete this training of the cell detection model.

[0006] Therefore, the sample image is subjected to target detection through the cell detection model, so as to obtain the detection result of the detection area including the target cells in the sample image, and then determine the loss value of the cell detection model based on the differences between the detection area and the actual area and the suspected area of ​​the target cells, and use the loss value to adjust the parameters of the cell detection model to complete the training of the cell detection model. In this way, the suspected area can be added during the iterative training of the cell detection model, and the unlabeled area outside the actual area in the sample image can be fully utilized, thereby improving the accuracy of cell detection.

[0007] Before determining the loss value of the cell detection model based on the differences between the detection area and the actual area and the suspected area of ​​the target cell, the method further includes: taking at least one detection area obtained in the previous training process as the suspected area of ​​the target cell.

[0008] Therefore, by using at least one detection area obtained in the previous training process as the suspected area of ​​the target cell, the suspected area of ​​the target cell can be detected by using the cell detection model during the training process, and this can be used to supervise the training of the cell detection model, which is beneficial to continuously improve the detection accuracy of the cell detection model during the iterative training process.

[0009] Wherein, the degree of overlap between at least one detection area and the actual area does not satisfy a preset condition.

[0010] Therefore, by setting at least one detection area so that the degree of overlap with the actual area does not satisfy a preset condition, suspected areas can be screened based on the degree of overlap between the detection area and the actual area, which can help reduce the possibility of missing suspected areas.

[0011] Among them, the overlap degree is the intersection-and-union ratio, and the preset conditions include: the intersection-and-union ratio is greater than or equal to the preset intersection-and-union ratio threshold; and / or, at least one detection area obtained in the previous training process is the detection area obtained in the previous training process, or is the detection area accumulated from several previous training processes.

[0012] Therefore, by taking the detection area with an IoU greater than or equal to a preset IoU threshold as the suspected area, it is possible to find the suspected areas that may be missing in the sample image, which can help reduce the probability of missed detection and improve the accuracy of cell detection; by setting at least one detection area obtained in the previous training process as the detection area obtained in the last training process, or setting it as the detection area accumulated from several previous training processes, it can help reduce the probability of missed detection and improve the accuracy of cell detection.

[0013] Among them, the detection result also includes the confidence corresponding to the detection area; based on the differences between the detection area and the actual area and the suspected area of ​​the target cell, the loss value of the cell detection model is determined, including: based on the difference between the detection area and the actual area, the actual loss value is obtained, based on the difference between the detection area and the suspected area, the suspected loss value is obtained; the actual loss value and the suspected loss value are weighted respectively to obtain the loss value of the cell detection model, wherein the weights of the actual loss value and the suspected loss value are determined by the confidence corresponding to the actual area and the suspected area, respectively.

[0014] Therefore, the actual loss value is obtained by detecting the difference between the area and the actual area, and the suspected loss value is obtained by detecting the difference between the area and the suspected area. The actual loss value and the suspected loss value are weighted respectively to obtain the loss value of the cell detection model. The weights of the actual loss value and the suspected loss value are determined by the confidence corresponding to the actual area and the suspected area, respectively. Therefore, during the training process, the actual area in the sample image and the suspected area with higher confidence can be fully learned, thereby increasing the probability of detecting target cells that may be missed in the sample image, thereby improving the accuracy of cell detection.

[0015] Among them, before using the cell detection model to perform target detection on the sample image and obtain the detection result, the method also includes at least one of the following steps: data enhancement on the sample image; color migration on the sample image; normalization of pixel values ​​in the sample image.

[0016] Therefore, by performing data enhancement on sample images, it can help avoid overfitting that may be caused by insufficient data, thereby helping to improve the accuracy of cell detection; by performing color migration on sample images, it can help adapt to the staining differences of multi-center data; by normalizing the pixel values ​​in the sample images, it can help to improve the convergence speed.

[0017] Wherein, the target cells include signet ring cells; and / or, the cell detection model includes at least one of a deformable convolutional layer and a global information enhancement module.

[0018] Therefore, by setting the target cells to include signet ring cells, the cell detection model can detect signet ring cells; by setting the cell detection model to include a deformable convolutional layer, the accuracy of identifying polymorphic target cells can be improved; by setting the cell detection model to include a global information enhancement module, it can be helpful to obtain long-distance, dependent features, which is beneficial to improving the accuracy of target cell identification.

[0019] The second aspect of the present application provides an image recognition method, comprising: obtaining a pathological image to be identified; using a cell detection model to perform target detection on the pathological image to be identified, and obtaining a detection area of ​​a target cell in the pathological image to be identified; wherein the cell detection model is obtained by the training method of the cell detection model in the first aspect mentioned above.

[0020] Therefore, by using the cell detection model obtained by the training method of the cell detection model in the first aspect above to perform target detection on the pathological image to be identified, the detection area of ​​the target cell in the pathological image to be identified is obtained, which can reduce the probability of missed detection and thus improve the accuracy of cell detection.

[0021] The third aspect of the present application provides an image recognition method, including: acquiring a pathological image to be identified; using a cell detection model to perform target detection on the pathological image to be identified, and obtaining a first detection area of ​​target cells in the pathological image to be identified; using a segmentation model to perform target segmentation on the pathological image to be identified, and obtaining a second detection area of ​​a preset target in the pathological image to be identified; screening the first detection area that meets the preset position condition between the first detection area and the second detection area as the final detection area of ​​the target cell.

[0022] Therefore, by using a cell detection model to perform target detection on the pathological image to be identified, a first detection area of ​​the target cells in the pathological image to be identified is obtained, and by using a segmentation model to perform target segmentation on the pathological image to be identified, a second detection area of ​​the preset target in the pathological image to be identified is obtained, thereby screening the first detection area that meets the preset position condition between the first detection area and the second detection area as the final detection area of ​​the target cells, which can help distinguish between target cells free from the preset target and normal cells located within the preset target, thereby reducing the probability of false detection and improving the accuracy of cell detection.

[0023] The preset position condition includes: the center of the first detection area is outside the second detection area; or the overlap between the first detection area and the second detection area is not greater than a preset threshold.

[0024] Therefore, by setting the preset position conditions to include: the center of the first detection area is outside the second detection area; or, the overlap between the first detection area and the second detection area is not greater than a preset threshold, it can be helpful to distinguish between target cells that are free from the preset target and normal cells that are within the preset target, thereby reducing the probability of false detection and improving the accuracy of cell detection.

[0025] Among them, the target cells include signet ring cells, and the preset targets include glands; and / or the cell detection model includes at least one of a deformable convolutional layer and a global information enhancement module; and / or the cell detection model is obtained by the training method of the cell detection model in the first aspect mentioned above; and / or the sample images used in the training of the segmentation model are processed by at least one of data enhancement, color migration, and pixel value normalization.

[0026] Therefore, by setting the target cells to include signet ring cells and the preset targets to include glands, the cell detection model can detect signet ring cells, and setting the preset target to glands can help distinguish signet ring cells free from glands; by setting the cell detection model to include a deformable convolution layer, the accuracy of identifying polymorphic target cells can be improved, and by setting the cell detection model to include a global information enhancement module, it can help obtain long-distance, dependent features, which can help improve the accuracy of target cell identification; by obtaining a cell detection model through the training method of the cell detection model in the first aspect above, the accuracy of cell detection can be further improved; by performing data enhancement on the sample image, it can help avoid overfitting that may be caused by insufficient data, thereby helping to improve the accuracy of cell detection; by performing color migration on the sample image, it can help adapt to the staining differences of multi-center data; by normalizing the pixel values ​​in the sample image, it can help to improve the convergence speed.

[0027] The fourth aspect of the present application provides a training device for a cell detection model, comprising: a target detection module, a loss determination module, a parameter adjustment module and a suspected search module, wherein the target detection module is used to perform target detection on a sample image using a cell detection model to obtain a detection result, wherein the detection result includes a detection area of ​​a target cell in the sample image, and the sample image is annotated with an actual area where the target cell is located; the loss determination module is used to determine the loss value of the cell detection model based on the differences between the detection area and the actual area and the suspected area of ​​the target cell; the parameter adjustment module is used to use the loss value to adjust the parameters of the detection model to complete the current training of the cell detection model.

[0028] In a fifth aspect, the present application provides an image recognition device, comprising: an image acquisition module and a target detection module, wherein the image acquisition module is used to acquire a pathological image to be identified; the target detection module is used to perform target detection on the pathological image to be identified using a cell detection model to obtain a detection area of ​​a target cell in the pathological image to be identified; wherein the cell detection model is obtained by a training device for the cell detection model in the fourth aspect.

[0029] In a sixth aspect, the present application provides an image recognition device, including: an image acquisition module, a target detection module, a target segmentation module and a region screening module, wherein the image acquisition module is used to acquire a pathological image to be identified; the target detection module is used to perform target detection on the pathological image to be identified using a cell detection model, and obtain a first detection region of target cells in the pathological image to be identified; the target segmentation module is used to perform target segmentation on the pathological image to be identified using a segmentation model, and obtain a second detection region of a preset target in the pathological image to be identified; the region screening module is used to screen a first detection region that meets a preset position condition between the first detection region and the second detection region, so as to serve as a final detection region of the target cell.

[0030] The seventh aspect of the present application provides an electronic device, comprising a memory and a processor coupled to each other, the processor being used to execute program instructions stored in the memory to implement the training method of the cell detection model in the first aspect above, or to implement the image recognition method in the second aspect above, or to implement the image recognition method in the third aspect above.

[0031] In an eighth aspect, the present application provides a computer-readable storage medium having program instructions stored thereon. When the program instructions are executed by a processor, the training method of the cell detection model in the first aspect mentioned above is implemented, or the image recognition method in the second aspect mentioned above is implemented, or the image recognition method in the third aspect mentioned above is implemented.

[0032] The above scheme can improve the accuracy of cell detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a flowchart of an embodiment of a training method for a cell detection model of the present application;

[0034] Figure 2 yes Figure 1 A schematic diagram of a flow chart of an embodiment of step S12;

[0035] Figure 3 It is a flowchart of an embodiment of the image recognition method of the present application;

[0036] Figure 4 yes Figure 3 A schematic diagram of an embodiment of a pathological image to be identified and its detection area;

[0037] Figure 5 is a flowchart of another embodiment of the image recognition method of the present application;

[0038] Figure 6 yes Figure 5 A schematic diagram of another embodiment of a pathological image to be identified and its detection area;

[0039] Figure 7is a schematic diagram of an embodiment of a final detection area after screening;

[0040] Figure 8 It is a schematic diagram of the framework of an embodiment of a training device for a cell detection model of the present application;

[0041] Fig. 9 It is a schematic diagram of the framework of an embodiment of the image recognition device of the present application;

[0042] Fig.10 is a schematic diagram of a framework of another embodiment of the image recognition device of the present application;

[0043] Fig.11 It is a schematic diagram of the framework of an embodiment of the electronic device of the present application;

[0044] Fig.12 It is a schematic diagram of a framework of an embodiment of a computer-readable storage medium of the present application. DETAILED DESCRIPTION

[0045] The scheme of the embodiment of the present application is described in detail below in conjunction with the drawings of the specification.

[0046] In the following description, for the purpose of explanation rather than limitation, specific details such as specific system structures, interfaces, and technologies are provided to facilitate a thorough understanding of the present application.

[0047] The terms "system" and "network" are often used interchangeably in this article. The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship. In addition, "many" in this article means two or more than two.

[0048] See also Figure 1 , Figure 1 This is a flow chart of an embodiment of a training method for a cell detection model of the present application. Specifically, the following steps may be included:

[0049] Step S11: Use the cell detection model to perform target detection on the sample image to obtain a detection result.

[0050] The cell detection model can adopt the Faster RCNN (Region with Convolutional Neural Networks) network model. In an implementation scenario, the cell detection model can also adopt Fast RCNN, YOLO (You Only Look Once), etc., which are not limited here. In a specific implementation scenario, when the cell detection model adopts Faster RCNN, ResNet101 can be used as its network for feature extraction. The sample image can be a pathological image, and the pathological image can include but is not limited to: cervical pathological images, liver pathological images, and kidney pathological images, which are not limited here.

[0051] The detection result includes the detection area of ​​the target cell in the sample image, and the sample image is annotated with the actual area where the target cell is located. The target cell can be set according to the detection task of the cell detection model. For example, the target cell can be a signet ring cell, so that the cell detection model can detect the signet ring cell after training. Other application scenarios can be deduced by analogy, and no examples are given here. The detection area can be represented by the center coordinates of a rectangle and the length and width of the rectangle. For example, (50, 60, 10, 20) can be used to represent a rectangle centered at the pixel point (50, 60) in the sample image, with a length of 10 and a width of 20. In an implementation scenario, the detection area can also use the center coordinates of a rectangle containing the target cell and the ratio of the length and width of the rectangle to the length and width of a preset rectangle. For example, the length of the preset rectangle is 10 and the width is 20, and (50, 60, 1, 1,) can be used to represent a rectangle centered at the pixel point (50, 60) in the sample image, with a length of 10 and a width of 20. In addition, the actual area can also be represented by the center coordinates of a rectangle containing the target cell and the length and width of the rectangle, which will not be given one by one here.

[0052] In one implementation scenario, before using the cell detection model to perform target detection on the sample image, the sample image can also be data enhanced. The data enhancement method may include but is not limited to: random rotation, random flipping, color perturbation, gamma correction, Gaussian noise, etc., so as to help avoid overfitting that may be caused by insufficient data. In another implementation scenario, before using the cell detection model to perform target detection on the sample image, the sample image can also be color migrated, so as to reduce the staining difference of multi-center data. In another implementation scenario, in order to improve the convergence speed, before using the cell detection model to perform target detection on the sample image, the pixel values ​​in the sample image can also be normalized. In a specific implementation scenario, the mean and variance of the image pixel values ​​in a preset large data set can be counted first, and then the pixel values ​​in each sample image are subtracted from the mean, and then divided by the variance, so as to normalize each sample image. The large data set can be Imagenet, PASCAL VOC, etc., which are not limited here.

[0053] In one implementation scenario, in order to improve the accuracy of identifying multi-morphological target cells, the cell detection model may include a deformable convolution layer, which is based on the position information of the space, and further displacement adjustment is made to achieve feature extraction of cells of different morphologies. In another implementation scenario, in order to obtain long-range, dependent features, thereby improving the accuracy of target cell identification, the cell detection model may also include a global information enhancement module.

[0054] Step S12: Determine the loss value of the cell detection model based on the differences between the detection area and the actual area and the suspected area of ​​the target cell.

[0055] In one implementation scenario, the suspected area of ​​the target cell can be an area suspected to be the target cell outlined by the user. In addition, during the outlining process, the user can also assign a confidence level to each suspected area, indicating the degree of certainty that the suspected area is indeed the target cell. For example, during the training process, the user's outlining information for the sample image can be received, and based on the outlining information, the suspected area of ​​the target cell can be obtained. In addition, based on the outlining information, the confidence level (for example, 90%) of the corresponding suspected area can be obtained. The above method can be beneficial to improving user perception through interaction with the user.

[0056] In another implementation scenario, the target cells in multiple sample images can also be counted to obtain the distribution law information of the target cells, so that the suspected area of ​​the target cells in the sample image can be obtained based on the distribution law information and the actual area marked in the sample image. In addition, a confidence level can be assigned to each suspected area, indicating the degree of certainty that the suspected area is indeed the target cell. For example, the distribution law information can include that there is at least one target cell within a preset range of each target cell. In this case, the actual area marked in the sample image can be analyzed, and if there is no other actual area within the preset range of the actual area, several suspected areas can be automatically set within the preset range of the actual area, and a confidence level (for example, 50%) can be assigned to each suspected area. The above method can avoid the need to manually outline the suspected area and improve the degree of automation.

[0057] In yet another implementation scenario, in order to improve the accuracy of the suspected region, at least one detection region obtained in the previous training process may also be used as the suspected region of the target cell.

[0058] In a specific implementation scenario, the overlap between at least one detection area and the actual area obtained in the above training process does not meet the preset conditions, that is, the overlap between the suspected area and the actual area does not meet the preset conditions. Specifically, the overlap can be the intersection over union (IoU), that is, the ratio between the pixels that exist in the detection area and the actual area at the same time, and the pixels that exist in the detection area or the actual area. The larger the IoU, the higher the overlap between the detection area and the actual area, and the smaller the IoU, the lower the overlap between the detection area and the actual area. In addition, the preset conditions may include: the IoU is greater than or equal to a preset IoU threshold (for example, 0.3, 0.4, etc.). By statistically analyzing the overlap between the detection area and the actual area during the training process, it is possible to find target cells that may be missed in the sample image.

[0059] In another specific implementation scenario, at least one detection area obtained in the previous training process may be a detection area obtained in the previous training process. For example, in the second training process, at least one detection area obtained in the first training process may be used as a suspected area in this training process; in the third training process, at least one detection area obtained in the second training process may be used as a suspected area in this training process, and so on. Examples are not given one by one here.

[0060] In another specific implementation scenario, at least one detection area obtained in the previous training process may also be the detection area accumulated from several previous training processes. For example, in the second training process, at least one detection area obtained in the first training process may be used as the suspected area of ​​this training process; in the third training process, at least one detection area obtained in the first and second training processes may be used as the suspected area of ​​this training process, and so on. Examples are not given one by one here.

[0061] In another specific implementation scenario, at least one detection area obtained by the above training process may not satisfy the preset condition with the center distance between the actual area, that is, the center distance between the suspected area and the actual area does not satisfy the preset condition. Specifically, the center distance may be the length of the line connecting the center of the detection area and the center of the actual area. In addition, the preset condition may include: the center distance is less than a preset center distance threshold. In the above method, the detection area with a larger center distance from the actual area can be used as a suspected area, so that target cells that may be missed in the sample image can be found.

[0062] In one implementation scenario, the difference between the detection area and the actual area and the suspected area may include: the difference in the center position of the area, the difference in the size of the area, etc., which is not limited here. Specifically, any one of the cross entropy loss value function and the mean square error function can be used to determine the loss value of the cell detection model in this training process. In one implementation scenario, the detection result may further include a confidence corresponding to the detection area, and the confidence indicates the credibility of the target cell contained in the detection area, that is, the higher the confidence, the greater the probability of the target cell contained in the detection area, and vice versa, the lower the confidence, the smaller the probability of the target cell contained in the detection area. The difference between the detection area and the actual area and the suspected area may include: the difference in the center position of the area, the difference in the size of the area, the difference in confidence, etc. In a specific implementation scenario, the confidence of the detection area can be obtained by cell detection model detection; in another specific implementation scenario, when the suspected area is obtained by at least one detection area, the confidence of the suspected area can be obtained by cell detection model detection. In addition, the confidence of the actual area can be set to 1.

[0063] Step S13: using the loss value, adjusting the parameters of the cell detection model to complete the current training of the cell detection model.

[0064] Specifically, the parameters of the cell detection model can be adjusted using loss values ​​by using methods such as stochastic gradient descent (SGD), batch gradient descent (BGD), and mini-batch gradient descent (MBGD). Batch gradient descent refers to using all samples to update parameters at each iteration; stochastic gradient descent refers to using one sample to update parameters at each iteration; and mini-batch gradient descent refers to using a batch of samples to update parameters at each iteration. Details will not be given here.

[0065] In an implementation scenario, a training end condition may also be set. When the training end condition is met, the training of the cell detection model may be terminated. Specifically, the training end condition may include: the loss value is less than a preset loss threshold, and the loss value no longer decreases; the current training times reaches a preset times threshold (for example, 500 times, 1000 times, etc.), which is not limited here.

[0066] The above scheme performs target detection on the sample image through the cell detection model, so as to obtain the detection result of the detection area including the target cells in the sample image, and then determines the loss value of the cell detection model based on the differences between the detection area and the actual area and the suspected area of ​​the target cell, and uses the loss value to adjust the parameters of the cell detection model to complete the current training of the cell detection model, so that the suspected area can be added in the iterative training process of the cell detection model, and the unlabeled area outside the actual area in the sample image can be fully utilized, thereby improving the accuracy of cell detection.

[0067] See also Figure 2 , Figure 2 yes Figure 1 Schematic diagram of the process of step S12 in the embodiment of the present disclosure. In the present disclosure embodiment, the suspected area of ​​the target cell is obtained by using at least one detection area obtained by previous training. For details, please refer to the relevant steps in the aforementioned disclosure embodiment, which will not be repeated here. In addition, the detection result may also include the confidence corresponding to the detection area, and determining the loss value of the cell detection model may specifically include the following steps:

[0068] Step S121: Based on the difference between the detected area and the actual area, the actual loss value is obtained, and based on the difference between the detected area and the suspected area, the suspected loss value is obtained.

[0069] In an implementation scenario, the actual loss value may include at least one of a region center position loss value, a region size loss value, and a region confidence loss value. The region center loss value may be calculated based on the difference between the center position of the detection region and the actual region, specifically, the cross entropy loss function may be used for calculation; the region size loss value may be calculated based on the difference between the region size (length, width) of the detection region and the actual region, specifically, the mean square error function may be used for calculation; the region confidence loss value may be calculated based on the confidence of the detection region and the actual region, specifically, the cross entropy loss function may be used for calculation, which will not be described in detail here.

[0070] In another implementation scenario, the suspected loss value may include at least one of a region center loss value, a region size loss value, and a region confidence loss value. The region center loss value may be calculated based on the difference between the center position of the detection region and the suspected region, specifically, the cross entropy loss function may be used for calculation; the region size loss value may be calculated based on the difference between the region size (length, width) of the detection region and the suspected region, specifically, the mean square error function may be used for calculation; the region confidence loss value may be calculated based on the confidence of the detection region and the suspected region, specifically, the cross entropy loss function may be used for calculation, which will not be described in detail here.

[0071] Step S122: performing weighted processing on the actual loss value and the suspected loss value respectively to obtain the loss value of the cell detection model.

[0072] Specifically, the weights of the actual loss value and the suspected loss value are determined by the confidence corresponding to the actual area and the suspected area, respectively. In a specific implementation scenario, the confidence of the actual area can be set to 1, and the loss value of the cell detection model can be expressed as:

[0073]

[0074] Among them, loss model represents the loss value of the cell detection model, represents the actual loss value corresponding to the ith actual area, represents the suspected loss value corresponding to the jth suspected area, λ j The confidence level corresponding to the jth suspected region.

[0075] Different from the previous embodiment, the actual loss value is obtained by detecting the difference between the area and the actual area, and the suspected loss value is obtained by detecting the difference between the area and the suspected area, so that the actual loss value and the suspected loss value are weighted respectively to obtain the loss value of the cell detection model, and the weights of the actual loss value and the suspected loss value are determined by the confidence corresponding to the actual area and the suspected area respectively. Therefore, during the training process, the actual area in the sample image and the suspected area with higher confidence can be fully learned, thereby increasing the probability of detecting target cells that may be missed in the sample image, thereby improving the accuracy of cell detection.

[0076] See also Figure 3 , Figure 3 This is a flow chart of an embodiment of the image recognition method of the present application. Specifically, it may include the following steps:

[0077] Step S31: Acquire a pathological image to be identified.

[0078] The pathological images to be identified may include, but are not limited to: cervical pathological images, liver pathological images, and kidney pathological images, which are not limited here.

[0079] Step S32: using the cell detection model to perform target detection on the pathological image to be identified, and obtaining a detection area of ​​the target cell in the pathological image to be identified.

[0080] Specifically, the cell detection model is trained by the steps in any of the above-mentioned cell detection model training method embodiments. For details, please refer to the relevant steps in the above-mentioned embodiments, which will not be repeated here. For example, if the cell detection model is trained using sample images labeled with signet ring cells, the cell detection model can be used to perform target detection on the pathological image to be identified, thereby obtaining the detection area of ​​signet ring cells in the pathological image to be identified. Please refer to Figure 4 , Figure 4 yes Figure 3 A schematic diagram of an embodiment of a pathological image to be identified and its detection area, as shown in FIG. Figure 4 As shown, Figure 4 The rectangular area with a black border in the middle represents the detection area detected by the cell detection model.

[0081] In one implementation scenario, after obtaining the detection area of ​​the target cells in the pathological image to be identified, the identified detection area can be statistically analyzed to obtain the number, size (such as maximum size, minimum size, average size, etc.), and shape of the target cells in the pathological image to be identified, which are not limited here.

[0082] The above scheme, by using the cell detection model obtained by the training method of the cell detection model in the above-mentioned first aspect to perform target detection on the pathological image to be identified, obtains the detection area of ​​the target cell in the pathological image to be identified, which can reduce the probability of missed detection, thereby improving the accuracy of cell detection.

[0083] See also Figure 5 , Figure 5 This is a flow chart of another embodiment of the image recognition method of the present application. Specifically, it may include the following steps:

[0084] Step S51: Acquire a pathological image to be identified.

[0085] The pathological images to be identified may include, but are not limited to: cervical pathological images, liver pathological images, and kidney pathological images, which are not limited here.

[0086] Step S52: using the cell detection model to perform target detection on the pathological image to be identified, and obtaining a first detection region of target cells in the pathological image to be identified.

[0087] In an implementation scenario, the cell detection model can be trained by the steps in any of the above-mentioned cell detection model training method embodiments, thereby further reducing the probability of missed detection by the cell detection model and improving the accuracy of cell detection. In addition, the cell detection model can also be obtained through the following training steps: using the cell detection model to perform target detection on the sample image to obtain the detection area of ​​the target cell, based on the difference between the detection area of ​​the target cell and the actual area marked in the sample image, determine the loss value of the cell detection model, and use the loss value to adjust the parameters of the cell detection model. Specifically, please refer to the relevant steps in the aforementioned disclosed embodiments, which will not be repeated here.

[0088] In one implementation scenario, in order to improve the accuracy of identifying polymorphic target cells, the cell detection model may include a deformable convolution layer, which makes further displacement adjustments based on the position information adopted in space to achieve feature extraction of cells with different morphologies.

[0089] In another implementation scenario, in order to obtain long-range, dependent features and thus improve the accuracy of target cell identification, the cell detection model may further include a global information enhancement module.

[0090] Please refer to Figure 4 ,like Figure 4 As shown, the cell detection model is used to perform target detection on the pathological image to be identified, and a first detection area (rectangular area with a black border) of the target cells in the pathological image to be identified is obtained.

[0091] Step S53: using the segmentation model to perform target segmentation on the pathological image to be identified, and obtaining a second detection area of ​​the preset target in the pathological image to be identified.

[0092] In one implementation scenario, the preset target may specifically be a gland, such as a breast, a prostate, etc., which is not limited here. In addition, the preset target may also include: a gastrointestinal tract, a bladder, etc.

[0093] In one implementation scenario, the segmentation model may include but is not limited to: Unet, FCN (Fully Convolutional Network), Enet, etc. In a specific implementation scenario, when Unet is used as the segmentation model, the network used for feature extraction may use DLAnet. In another specific implementation scenario, the samples used in the training of the segmentation model may be processed by at least one of data enhancement, color migration, and pixel value normalization, wherein data enhancement, color migration, and pixel value normalization may refer to the relevant steps in the aforementioned embodiment, which will not be repeated here. In another implementation scenario, the segmentation model and the cell detection model may be trained with the same sample images during the training process, or may be trained with different sample images, which is not limited here. In another implementation scenario, the segmentation model and the cell detection model may be trained in a sequential order, for example, the cell detection model may be trained first, and then the segmentation model, or the segmentation model may be trained first, and then the cell detection model; the segmentation model and the cell detection model may also be trained simultaneously, which is not limited here.

[0094] Please refer to Figure 6 , Figure 6 yes Figure 5 Schematic diagram of another embodiment of a pathological image to be identified and its detection area. Figure 6 As shown, Figure 6 The area circled by the thick black line represents the second detection area of ​​the preset target.

[0095] The above steps S52 and S53 can be performed in a sequential order, for example, step S52 can be performed first and then step S53; or step S53 can be performed first and then step S52. The above steps S52 and S53 can also be performed simultaneously, which is not limited here.

[0096] Step S54: selecting the first detection area that meets the preset position condition between the first detection area and the second detection area as the final detection area of ​​the target cells.

[0097] In one implementation scenario, the preset location condition may include that the center of the first detection area is outside the second detection area, for example, Figure 4If the center of the first detection area (the rectangular area with a black border) is Figure 6 If the second detection area (the area circled by the thick black line) is outside the second detection area shown in FIG, it can be used as the final detection area of ​​the target cells. Figure 7 , Figure 7 is a schematic diagram of an embodiment of the final detection area after screening, such as Figure 7 As shown, the first detection area whose center is outside the second detection area is screened to obtain a final detection area (a rectangular area with a black frame in the figure), thereby identifying the target cells in the pathological image to be identified.

[0098] In another implementation scenario, the preset location condition may also include that the overlap between the first detection area and the second detection area is not greater than a preset threshold. In a specific implementation scenario, the overlap may be the intersection over union (IoU). In this case, the preset threshold may be a preset intersection over union threshold (for example, 0.8, 0.9, etc.).

[0099] In one implementation scenario, after obtaining the final detection area of ​​the target cells in the pathological image to be identified, the identified final detection area can be statistically analyzed to obtain the number, size (such as maximum size, minimum size, average size, etc.), and shape of the target cells in the pathological image to be identified, which are not limited here.

[0100] The above scheme uses a cell detection model to perform target detection on the pathological image to be identified, obtains a first detection area of ​​the target cells in the pathological image to be identified, and uses a segmentation model to perform target segmentation on the pathological image to be identified, obtains a second detection area of ​​the preset target in the pathological image to be identified, thereby screening the first detection area that meets the preset position condition between the first detection area and the second detection area as the final detection area of ​​the target cells, which can help distinguish between target cells free from the preset target and normal cells located within the preset target, thereby reducing the probability of false detection and improving the accuracy of cell detection.

[0101] See also Figure 8 , Figure 8It is a schematic diagram of the framework of an embodiment of a training device 80 of a cell detection model of the present application. The training device 80 of the cell detection model includes a target detection module 81, a loss determination module 82, and a parameter adjustment module 83. The target detection module 81 is used to perform target detection on a sample image using a cell detection model to obtain a detection result, wherein the detection result includes a detection area of ​​a target cell in the sample image, and the sample image is annotated with an actual area where the target cell is located; the loss determination module 82 is used to determine the loss value of the cell detection model based on the difference between the detection area and the actual area and the suspected area of ​​the target cell; the parameter adjustment module 83 is used to use the loss value to adjust the parameters of the detection model to complete the current training of the cell detection model.

[0102] The above scheme performs target detection on the sample image through the cell detection model, so as to obtain the detection result of the detection area including the target cells in the sample image, and then determines the loss value of the cell detection model based on the differences between the detection area and the actual area and the suspected area of ​​the target cell, and uses the loss value to adjust the parameters of the cell detection model to complete the current training of the cell detection model, so that the suspected area can be added in the iterative training process of the cell detection model, and the unlabeled area outside the actual area in the sample image can be fully utilized, thereby improving the accuracy of cell detection.

[0103] In some disclosed embodiments, the cell detection model training device 80 further includes a suspected search module 84 for using at least one detection region obtained in a previous training process as a suspected region of a target cell.

[0104] Different from the aforementioned embodiment, by using at least one detection area obtained in the previous training process as the suspected area of ​​the target cell, the suspected area of ​​the target cell can be detected by using the cell detection model during the training process, and this can be used to supervise the training of the cell detection model, which is beneficial to continuously improve the detection accuracy of the cell detection model during the iterative training process.

[0105] In some disclosed embodiments, a degree of overlap between at least one detection area and an actual area does not satisfy a preset condition.

[0106] Different from the aforementioned embodiment, at least one detection area is set to have a degree of overlap with the actual area that does not satisfy a preset condition, so that suspected areas can be screened based on the degree of overlap between the detection area and the actual area, thereby helping to reduce the possibility of missing suspected areas.

[0107] In some disclosed embodiments, the degree of overlap is an intersection-and-union ratio, and preset conditions include: the intersection-and-union ratio is greater than or equal to a preset intersection-and-union ratio threshold; and / or, at least one detection area obtained in a previous training process is a detection area obtained in a previous training process, or is a detection area accumulated from several previous training processes.

[0108] Different from the aforementioned embodiment, by taking the detection area whose IoU is greater than or equal to the preset IoU threshold as the suspected area, it is possible to find the suspected area that may be missing in the sample image, which can help reduce the probability of missed detection and improve the accuracy of cell detection; by setting at least one detection area obtained in the previous training process as the detection area obtained in the last training process, or setting it as the detection area accumulated from several previous training processes, it can help reduce the probability of missed detection and improve the accuracy of cell detection.

[0109] In some disclosed embodiments, the detection result also includes a confidence level corresponding to the detection area. The loss determination module 82 includes a loss value determination submodule for obtaining an actual loss value based on a difference between the detection area and the actual area, and obtaining a suspected loss value based on a difference between the detection area and the suspected area. The loss determination module 82 also includes a loss weighting submodule for weighting the actual loss value and the suspected loss value, respectively, to obtain a loss value of the cell detection model, wherein the weights of the actual loss value and the suspected loss value are determined by the confidence levels corresponding to the actual area and the suspected area, respectively.

[0110] Different from the previous embodiment, the actual loss value is obtained by detecting the difference between the area and the actual area, and the suspected loss value is obtained by detecting the difference between the area and the suspected area, so that the actual loss value and the suspected loss value are weighted respectively to obtain the loss value of the cell detection model, and the weights of the actual loss value and the suspected loss value are determined by the confidence corresponding to the actual area and the suspected area respectively. Therefore, during the training process, the actual area in the sample image and the suspected area with higher confidence can be fully learned, thereby increasing the probability of detecting target cells that may be missed in the sample image, thereby improving the accuracy of cell detection.

[0111] In some disclosed embodiments, the training device 80 for the cell detection model also includes a data enhancement module for performing data enhancement on the sample image, the training device 80 for the cell detection model also includes a color migration module for performing color migration on the sample image, and the training device 80 for the cell detection model also includes a normalization module for normalizing the pixel values ​​in the sample image.

[0112] Different from the aforementioned embodiments, by performing data enhancement on the sample image, it can help avoid overfitting that may be caused by insufficient data, thereby helping to improve the accuracy of cell detection; by performing color migration on the sample image, it can help adapt to the staining differences of multi-center data; by normalizing the pixel values ​​in the sample image, it can help to improve the convergence speed.

[0113] In some disclosed embodiments, the target cell comprises a signet ring cell; and / or the cell detection model comprises at least one of a deformable convolutional layer and a global information enhancement module.

[0114] Different from the aforementioned embodiments, by setting the target cells to include signet ring cells, the cell detection model can detect signet ring cells; by setting the cell detection model to include a deformable convolutional layer, the accuracy of identifying polymorphic target cells can be improved; by setting the cell detection model to include a global information enhancement module, it can be helpful to obtain long-distance, dependent features, which is beneficial to improving the accuracy of target cell identification.

[0115] See also Fig. 9 , Fig. 9 1 is a schematic diagram of a framework of an embodiment of an image recognition device 90 of the present application. The image recognition device 90 includes an image acquisition module 91 and a target detection module 92. The image acquisition module 91 is used to acquire a pathological image to be identified; the target detection module 92 is used to perform target detection on the pathological image to be identified using a cell detection model to obtain a detection area of ​​a target cell in the pathological image to be identified; wherein the cell detection model is trained by the cell detection model training device in any of the above cell detection model training device embodiments.

[0116] The above scheme, by using the cell detection model obtained by the cell detection model training device in any of the above-mentioned cell detection model training device embodiments to perform target detection on the pathological image to be identified, obtains the detection area of ​​the target cell in the pathological image to be identified, which can reduce the probability of missed detection, thereby improving the accuracy of cell detection.

[0117] See also Fig.10 , Fig.10It is a schematic diagram of the framework of an embodiment of the image recognition device 100 of the present application. The image recognition device 100 includes an image acquisition module 101, a target detection module 102, a target segmentation module 103 and a region screening module 104. The image acquisition module 101 is used to acquire the pathological image to be identified; the target detection module 102 is used to use the cell detection model to perform target detection on the pathological image to be identified, and obtain the first detection area of ​​the target cell in the pathological image to be identified; the target segmentation module 103 is used to use the segmentation model to perform target segmentation on the pathological image to be identified, and obtain the second detection area of ​​the preset target in the pathological image to be identified; the region screening module 104 is used to screen the first detection area that meets the preset position condition between the first detection area and the second detection area, so as to serve as the final detection area of ​​the target cell.

[0118] The above scheme uses a cell detection model to perform target detection on the pathological image to be identified, obtains a first detection area of ​​the target cells in the pathological image to be identified, and uses a segmentation model to perform target segmentation on the pathological image to be identified, obtains a second detection area of ​​the preset target in the pathological image to be identified, thereby screening the first detection area that meets the preset position condition between the first detection area and the second detection area as the final detection area of ​​the target cells, which can help distinguish between target cells free from the preset target and normal cells located within the preset target, thereby reducing the probability of false detection and improving the accuracy of cell detection.

[0119] In some disclosed embodiments, the preset position condition includes: the center of the first detection area is outside the second detection area; or, the overlap between the first detection area and the second detection area is not greater than a preset threshold.

[0120] Different from the aforementioned embodiment, by setting the preset position condition to include: the center of the first detection area is outside the second detection area; or, the overlap between the first detection area and the second detection area is not greater than a preset threshold, it can be helpful to distinguish between target cells that are free from the preset target and normal cells that are within the preset target, thereby reducing the probability of false detection and improving the accuracy of cell detection.

[0121] In some disclosed embodiments, the target cells include signet ring cells, and the preset targets include glands; and / or the cell detection model includes at least one of a deformable convolutional layer and a global information enhancement module; and / or the cell detection model is trained by a cell detection model training device in any of the above-mentioned cell detection model training device embodiments; and / or the sample images used in the training of the segmentation model are processed by at least one of data enhancement, color migration, and pixel value normalization.

[0122] Different from the aforementioned embodiments, by setting the target cells to include signet ring cells, the cell detection model can detect signet ring cells, and by setting the preset target to the gland, it can be helpful to distinguish the signet ring cells free from the gland; by setting the cell detection model to include a deformable convolution layer, the accuracy of identifying polymorphic target cells can be improved, and by setting the cell detection model to include a global information enhancement module, it can be helpful to obtain long-distance, dependent features, which is helpful to improve the accuracy of target cell identification; by training the cell detection model with the cell detection model training device in any of the above-mentioned cell detection model training device embodiments to obtain a cell detection model, the accuracy of cell detection can be further improved; by performing data enhancement on the sample image, it can be helpful to avoid overfitting that may be caused by insufficient data, thereby helping to improve the accuracy of cell detection; by performing color migration on the sample image, it can be helpful to adapt to the staining differences of multi-center data; by normalizing the pixel values ​​in the sample image, it can be helpful to improve the convergence speed.

[0123] See also Fig.11 , Fig.11 1 is a schematic diagram of a framework of an embodiment of an electronic device 110 of the present application. The electronic device 110 includes a memory 111 and a processor 112 coupled to each other, and the processor 112 is used to execute program instructions stored in the memory 111 to implement the steps in the training method embodiment of any of the above-mentioned cell detection models, or to implement the steps in the embodiment of any of the above-mentioned image recognition methods. In a specific implementation scenario, the electronic device 110 may include, but is not limited to: a microcomputer, a server, and in addition, the electronic device 110 may also include mobile devices such as laptops and tablet computers, which are not limited here.

[0124] Specifically, the processor 112 is used to control itself and the memory 111 to implement the steps in the training method embodiment of any of the above-mentioned cell detection models, or to implement the steps in any of the above-mentioned image recognition method embodiments. The processor 112 can also be referred to as a CPU (Central Processing Unit). The processor 112 may be an integrated circuit chip with signal processing capabilities. The processor 112 can also be a general-purpose processor, a digital signal processor (DigitalSignal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. In addition, the processor 112 can be implemented by an integrated circuit chip.

[0125] The above scheme can improve the accuracy of cell detection.

[0126] See also Fig.12 , Fig.12 1 is a schematic diagram of a framework of an embodiment of a computer-readable storage medium 120 of the present application. The computer-readable storage medium 120 stores program instructions 121 that can be executed by a processor, and the program instructions 121 are used to implement the steps in any of the above-mentioned cell detection model training method embodiments, or to implement the steps in any of the above-mentioned image recognition method embodiments.

[0127] The above scheme can improve the accuracy of cell detection.

[0128] In the several embodiments provided in the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0129] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0130] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0131] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

Claims

1. A method for training a cell detection model, characterized in that: include: Performing target detection on the sample image using the cell detection model to obtain a detection result, wherein the detection result includes a detection area of ​​the target cell in the sample image, and the sample image is annotated with an actual area where the target cell is located; Determining a loss value of the cell detection model based on differences between the detection area and the actual area and the suspected area of ​​the target cell; Using the loss value, adjusting the parameters of the cell detection model to complete the current training of the cell detection model; Wherein, before determining the loss value of the cell detection model based on the differences between the detection area and the actual area and the suspected area of ​​the target cell, the method further includes: using at least one detection area obtained in a previous training process as the suspected area of ​​the target cell, and the degree of overlap between the at least one detection area and the actual area does not meet a preset condition.

2. The training method according to claim 1, characterized in that: The overlap degree is an intersection-and-union ratio, and the preset conditions include: the intersection-and-union ratio is greater than or equal to a preset intersection-and-union ratio threshold; And / or, the at least one detection area obtained in the previous training process is a detection area obtained in the last training process, or is a detection area accumulated from several previous training processes.

3. The training method according to claim 1 or 2, characterized in that: The detection result also includes a confidence level corresponding to the detection area; The determining the loss value of the cell detection model based on the differences between the detection area and the actual area and the suspected area of ​​the target cell respectively includes: Based on the difference between the detection area and the actual area, an actual loss value is obtained, and based on the difference between the detection area and the suspected area, a suspected loss value is obtained; The actual loss value and the suspected loss value are weighted respectively to obtain the loss value of the cell detection model, wherein the weights of the actual loss value and the suspected loss value are determined by the confidence corresponding to the actual area and the suspected area respectively.

4. The training method according to claim 1 or 2, characterized in that: Before performing target detection on the sample image using the cell detection model to obtain the detection result, the method further includes at least one of the following steps: Performing data enhancement on the sample image; Performing color migration on the sample image; The pixel values ​​in the sample image are normalized.

5. The training method according to claim 1 or 2, characterized in that: The target cells include signet ring cells; And / or, the cell detection model includes at least one of a deformable convolution layer and a global information enhancement module.

6. An image recognition method, characterized in that: include: Acquire a pathological image to be identified; Using a cell detection model to perform target detection on the pathological image to be identified, and obtaining a detection area of ​​a target cell in the pathological image to be identified; Wherein, the cell detection model is trained by the method described in any one of claims 1 to 5.

7. An image recognition method, characterized in that: include: Acquire a pathological image to be identified; Using a cell detection model to perform target detection on the pathological image to be identified, and obtaining a first detection area of ​​the target cell in the pathological image to be identified; wherein the cell detection model is trained by the method according to any one of claims 1 to 5; Performing target segmentation on the pathological image to be identified by using the segmentation model to obtain a second detection area of ​​a preset target in the pathological image to be identified; A first detection area that meets a preset position condition between the first detection area and the second detection area is selected as a final detection area for the target cells.

8. The image recognition method according to claim 7, characterized in that: The preset position condition includes: the center of the first detection area is outside the second detection area; or, the overlap between the first detection area and the second detection area is not greater than a preset threshold.

9. The image recognition method according to claim 7 or 8, characterized in that: The target cells include signet ring cells, and the preset targets include glands; And / or, the cell detection model includes at least one of a deformable convolution layer and a global information enhancement module; And / or, the sample images used in training the segmentation model are processed by at least one of data enhancement, color migration, and pixel value normalization.

10. A training device for a cell detection model, characterized in that: include: A target detection module, used to perform target detection on the sample image using the cell detection model to obtain a detection result, wherein the detection result includes a detection area of ​​the target cell in the sample image, and the sample image is annotated with an actual area where the target cell is located; A loss determination module, used to determine the loss value of the cell detection model based on the differences between the detection area and the actual area and the suspected area of ​​the target cell; A parameter adjustment module, used to adjust the parameters of the cell detection model using the loss value to complete the current training of the cell detection model; Among them, the training device of the cell detection model also includes a suspected search module, which is used to use at least one detection area obtained in the previous training process as the suspected area of ​​the target cell, and the overlap between the at least one detection area and the actual area does not meet the preset conditions.

11. An image recognition device, characterized in that: include: An image acquisition module, used for acquiring a pathological image to be identified; A target detection module, used to perform target detection on the pathological image to be identified using a cell detection model, and obtain a detection area of ​​a target cell in the pathological image to be identified; Wherein, the cell detection model is obtained by the training device of the cell detection model described in claim 10.

12. An image recognition device, characterized in that: include: An image acquisition module, used for acquiring a pathological image to be identified; A target detection module, used for performing target detection on the pathological image to be identified by using a cell detection model to obtain a first detection area of ​​target cells in the pathological image to be identified; wherein the cell detection model is obtained by the training device of the cell detection model according to claim 10; A target segmentation module, used for performing target segmentation on the pathological image to be identified by using a segmentation model to obtain a second detection area of ​​a preset target in the pathological image to be identified; The area screening module is used to screen the first detection area that meets the preset position condition between the first detection area and the second detection area to serve as the final detection area of ​​the target cell.

13. An electronic device, characterized in that: It comprises a memory and a processor coupled to each other, wherein the processor is used to execute program instructions stored in the memory to implement the training method of the cell detection model described in any one of claims 1 to 5, or the image recognition method described in any one of claims 6 to 9.

14. A computer-readable storage medium having program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the training method of the cell detection model described in any one of claims 1 to 5, or the image recognition method described in any one of claims 6 to 9 is implemented.

Images