A method, device and electronic device for identifying cells in cell clusters

Abstract

The present application provides a method, device and electronic device for identifying cells in cell clusters. The method for identifying cells in cell clusters includes: identifying at least one cell cluster region image from a digital slice image to be detected; for each cell cluster region image, identifying each single cell region image and a standard cell image from the cell cluster region image; for each single cell region image in each cell cluster region image, comparing the single cell region image with the standard cell image to determine whether the single cell in the single cell region image is a suspicious cell; inputting the suspicious cell image corresponding to each suspicious cell into a trained cell classification model to determine whether there is a target cell of the target type in the digital slice image. The present application can improve the accuracy of identifying target cells in cell clusters.

Description

Technical Field

[0001] The present application relates to the field of cell identification technology, and in particular to a method, device and electronic equipment suitable for identifying cells in a cell cluster. Background Art

[0002] With the rapid development of the biological industry, cells, as an important component of living organisms, have attracted more and more people to study cells. However, in the existing field of cell identification, most of the research and identification are aimed at single cells. However, in addition to free single cells, there are also a large number of cell clusters in living organisms. There is little analysis of cell clusters in the existing technology.

[0003] A cell cluster is a mass of tissue that gathers a large number of cells. Because a cell cluster gathers a large number of cells, the staining is darker, the overlap is serious, and the cell nucleus information is unclear. Therefore, it is impossible to accurately identify the cell type in the cell cluster. Summary of the invention

[0004] In view of this, the purpose of the present application is to provide a method, device and electronic equipment suitable for identifying cells in a cell cluster, which can improve the accuracy of identifying target cells in a cell cluster.

[0005] The present application embodiment provides a method for identifying cells in a cell cluster, and the method for identifying cells in a cell cluster includes:

[0006] identifying at least one cell cluster region image from the digital slice image to be detected;

[0007] For each of the cell cluster region images, identifying individual single cell region images and a standard cell image from the cell cluster region image;

[0008] For each of the single cell region images in each of the cell cluster region images, the single cell region image is compared with the standard cell image to determine whether the single cell in the single cell region image is a suspicious cell;

[0009] The suspicious cell images corresponding to the suspicious cells are input into the trained cell classification model to determine whether there are target cells of the target type in the digital slice image.

[0010] Furthermore, for each of the cell cluster region images, identifying individual single cell region images from the cell cluster region image includes:

[0011] For each of the cell cluster region images, identifying each cell nucleus from the cell cluster region image, and determining the cell image corresponding to each cell nucleus as a single cell region image;

[0012] Determine whether the target distance between any two single cell region images in the cell cluster region image is greater than or equal to a preset threshold;

[0013] If it is greater than, the target area image between any two single cell area images in the cell cluster area image is determined as a single cell area image.

[0014] Furthermore, for each of the single cell region images in each of the cell cluster region images, the single cell region image is compared with the standard cell image to determine whether the single cell in the single cell region image is a suspicious cell, including:

[0015] For each of the single cell region images in each of the cell cluster region images, the single cell region image and the standard cell image are input into a trained cell screening model to determine the similarity between each single cell in the cell cluster region image and the standard cell;

[0016] If the similarity is less than a preset threshold, the single cell corresponding to the similarity is determined as a suspicious cell in the cell cluster region image.

[0017] Further, the trained cell screening model is determined by the following methods:

[0018] Obtaining different types of sample single cell area images in the sample digital slice image, type labels of each of the sample single cell area images, and sample standard cell images corresponding to the sample digital slice image; the type label is used to characterize the real sample similarity between the sample single cell and the preset sample standard cell;

[0019] Inputting the sample single cell region image and the label of the sample single cell region image into an initial cell screening model to determine a preset sample similarity between the sample single cell and a sample standard cell;

[0020] When the loss value between the preset sample similarity and the real sample similarity between the sample single cell and the sample standard cell is less than a preset threshold, the training is terminated and the trained cell screening model is determined.

[0021] Furthermore, the loss value between the preset sample similarity and the real sample similarity is determined by the following formula:

[0022] Y=|A|,A<0;

[0023] Y=0,A>0;

[0024] in,

[0025] Wherein, Y is used to represent the loss value between the preset sample similarity and the real sample similarity, i is used to represent the number of feature vectors of the sample single cell, and j is used to represent the number of feature vectors of the sample standard cell; v bj The feature vector used to characterize the sample standard cell; v ai and v ci The feature vectors used to characterize the first type of sample single cells and the feature vectors used to characterize the second type of sample single cells respectively; N is used to characterize the number of sample single cells of the first type; 3N is the number of sample single cells of the second type.

[0026] The present application also provides an identification device for cells in a cell cluster. The identification device for cells in a cell cluster includes:

[0027] A first recognition module is used to recognize at least one cell cluster region image from the digital slice image to be detected;

[0028] A second recognition module is used to recognize, for each of the cell cluster region images, individual single cell region images and a standard cell image from the cell cluster region image;

[0029] A first determination module is used to compare each of the single cell region images in each of the cell cluster region images with the standard cell image to determine whether the single cell in the single cell region image is a suspicious cell;

[0030] The second determination module is used to input the suspicious cell images corresponding to the suspicious cells into the trained cell classification model to determine whether there are target cells of the target type in the digital slice image.

[0031] Furthermore, the second identification module is specifically used to:

[0032] For each of the cell cluster region images, identifying each cell nucleus from the cell cluster region image, and determining the cell image corresponding to each cell nucleus as a single cell region image;

[0033] Determine whether the target distance between any two single cell region images in the cell cluster region image is greater than or equal to a preset threshold;

[0034] If it is greater than, the target area image between any two single cell area images in the cell cluster area image is determined as a single cell area image.

[0035] Furthermore, the first determining module is specifically configured to:

[0036] For each of the single cell region images in each of the cell cluster region images, the single cell region image and the standard cell image are input into a trained cell screening model to determine the similarity between each single cell in the cell cluster region image and the standard cell;

[0037] If the similarity is less than a preset threshold, the single cell corresponding to the similarity is determined as a suspicious cell in the cell cluster region image.

[0038] An embodiment of the present application also provides an electronic device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the above-mentioned method for identifying cells in a cell cluster are performed.

[0039] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for identifying cells in a cell cluster are executed.

[0040] The methods, devices and electronic equipment for identifying cells in cell clusters provided in the embodiments of the present application, compared with the prior art, can improve the accuracy of identifying target cells in cell clusters by comparing the single-cell region image in the digital slice image to be detected with the standard cell image, and further accurately identifying the suspicious cells after the comparison, thereby determining the target type of the target cell in the digital slice image.

[0041] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 One of the flow charts of a method for identifying cells in a cell cluster provided in an embodiment of the present application is shown;

[0044] Figure 2 A structural diagram of a single cell region image in a method for identifying cells in a cell cluster provided in an embodiment of the present application is shown;

[0045] Figure 3 A second flowchart of another method for identifying cells in a cell cluster provided in an embodiment of the present application is shown;

[0046] Figure 4 A flow chart of an embodiment of a method for identifying cells in a cell cluster provided in an embodiment of the present application is shown;

[0047] Figure 5 A schematic diagram of the structure of a device for identifying cells in a cell cluster provided in an embodiment of the present application is shown;

[0048] Figure 6 A schematic structural diagram of an electronic device provided in an embodiment of the present application is shown.

[0049] In the figure:

[0050] 500 - an identification device for cells in a cell cluster; 510 - a first identification module; 520 - a second identification module; 550 - a first determination module; 540 - a second determination module; 600 - an electronic device; 610 - a processor; 620 - a memory; 630 - a bus. DETAILED DESCRIPTION

[0051] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application usually described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work belongs to the scope of protection of the present application.

[0052] First, the application scenarios to which this application is applicable are introduced. It has been found through research that in the prior art, a cell cluster is a mass tissue that gathers a large number of cells, and it is precisely because a cell cluster gathers a large number of cells that the staining is darker, the overlap is serious, and the cell nucleus information is unclear. Therefore, it is impossible to accurately identify the cell type in the cell cluster.

[0053] Based on this, the embodiments of the present application provide a method, device and electronic equipment suitable for identifying cells in a cell cluster, which can improve the accuracy of identifying target cells in a cell cluster.

[0054] See also Figure 1 , Figure 1This is a flow chart of one of the methods for identifying cells in a cell cluster provided in an embodiment of the present application. Figure 1 As shown in, the identification method for cells in a cell cluster provided in the embodiment of the present application includes:

[0055] S101 . Identify at least one cell cluster region image from a digital slice image to be detected.

[0056] In this step, the slice image to be detected is first decoded and digitally converted through the corresponding decoding program to generate a digital slice image to be detected, and then the digital slice image to be detected is processed according to multiple resolutions to determine the target digital slice image corresponding to each resolution, and then the digital slice image to be detected corresponding to each resolution is respectively input into the trained cell cluster detection model at the corresponding resolution to determine the cell cluster detection result at each resolution, and the cell cluster detection results at multiple resolutions are remapped to the maximum resolution, and the cell cluster detection results at each resolution are merged to identify at least one cell cluster area image.

[0057] Among them, the cell clusters in the cell cluster area image can be cell clusters composed of normal cell structures or cell clusters composed of atrophic cell structures. Since atrophic cell clusters often gather a large number of atrophic cells, are darkly stained, and have serious overlap, it is difficult to accurately identify target cells of the target type therefrom. Therefore, the method for identifying cells in cell clusters provided in the present application is used to identify target cells of the target type.

[0058] Here, the division of resolution can be customized according to the actual reference scenario. The multiple resolutions in the embodiments provided in this application can be 40 times, 20 times and 10 times. It can be seen that the maximum resolution in the embodiments provided in this application is 40 times.

[0059] The cell cluster detection results at each resolution include but are not limited to the position and size relationship of the cell cluster; and the method of merging the cell cluster detection results at each resolution includes but is not limited to merging using a non-maximum suppression method.

[0060] In this way, in the embodiment provided in the present application, the digital slice image to be detected with a resolution of 40 times can be specifically set to a digital slice image to be detected with a resolution of 20000x20000, the digital slice image to be detected with a resolution of 20 times can be specifically set to a digital slice image to be detected with a resolution of 10000x10000, and the digital slice image to be detected with a resolution of 10 times can be specifically set to a digital slice image to be detected with a resolution of 5000x5000.

[0061] Here, the cell cluster detection model trained at each resolution is determined in the following way:

[0062] The sample digital slice images at various resolutions are cut according to a preset size (eg, 512x512), and significant cell cluster regions are marked on each cut image to determine the sample cell cluster image.

[0063] The sample cell cluster images are input into the initial cell cluster detection model for training, and the trained cell cluster detection models at various resolutions are determined.

[0064] The initial cell cluster detection model may adopt but is not limited to detection network models such as Faster RCNN and SSD series.

[0065] In the above, the detection network models such as Faster RCNN and SSD series require that the sample cell cluster image at the corresponding resolution cannot exceed the standard image size (such as 256x256).

[0066] S102 . For each of the cell cluster region images, identify individual single cell region images and a standard cell image from the cell cluster region image.

[0067] In this step, for each of the cell cluster region images, the trained cell segmentation model is used to identify individual single cell region images from the cell cluster region image, and for each of the cell cluster region images, a customized standard cell image is labeled, and the customized standard cell image is a cell of the same cell type.

[0068] The trained cell segmentation model may be, but is not limited to, network segmentation models such as Mask-RCNN, Unet, and Deeplab series.

[0069] In this way, the trained cell segmentation model is determined by the following steps:

[0070] The sample cell cluster image is labeled to determine a sample single cell region image in the sample cell cluster image.

[0071] The sample single cell region image and the label of the sample single cell region image are input into an initial cell segmentation model, the initial cell segmentation model is trained, and each sample cell nucleus in the sample cell cluster image is determined.

[0072] In the above, the label of the sample single cell area image is used to characterize the position information and area size of the sample single cell area image, and the output of the position information and area size of each sample cell nucleus is used as the initial cell segmentation model in the training process to obtain the preset position information and preset area size of the sample single cell area image.

[0073] The preset position information and the preset area size are trained with the labels of the real annotated sample single-cell area images for loss value. When the loss value is less than the preset threshold, the training is terminated and the trained cell segmentation model is determined.

[0074] Among them, in the actual scene application process, the staff will use the initial cell segmentation model with different network parameters for model training, and select the cell segmentation model with the best training effect as the final trained cell segmentation model.

[0075] In step S102, for each of the cell cluster region images, identifying individual single cell region images from the cell cluster region image includes:

[0076] Step 1021: for each of the cell cluster region images, identify each cell nucleus from the cell cluster region image, and determine the cell image corresponding to each cell nucleus as a single cell region image.

[0077] For each of the cell cluster region images, each cell nucleus in each of the cell cluster region images is determined by using a trained cell segmentation model, and the cell image corresponding to each cell nucleus is determined as a single cell region image.

[0078] In this way, the cell image corresponding to each cell nucleus includes the position information corresponding to the cell nucleus and the size of the region dividing the cell nucleus.

[0079] Step 1022: determine whether the target distance between any two single cell region images in the cell cluster region image is greater than or equal to a preset threshold.

[0080] Among them, the size of the preset threshold can be customized and set according to the actual situation of different application scenarios, such as according to the slice size of the digital slice image to be detected. Here, the preset threshold in the embodiment provided in the present application can be set to the size of 1.5 single-cell area images.

[0081] Step 1023: If it is greater than, the target region image between any two of the single cell region images in the cell cluster region image is determined as a single cell region image.

[0082] Among them, Figure 2 As shown, Figure 2 This is a structural diagram of a single cell region image in a method for identifying cells in a cell cluster provided in an embodiment of the present application. Here, Figure 2What is shown in the figure is that when it is determined that the target distance between any two single-cell region images in the cell cluster region image is greater than or equal to a preset threshold, the target region between the any two single-cell region images in the cell cluster region image is also determined as a single-cell region image. At this time, the single-cell region image includes, in addition to the stored single-cell region image in the cell cluster region image, an image corresponding to the target region.

[0083] In this way, the purpose of determining the target area as a single-cell area image is to prevent the occurrence of single cell confirmation and omission when segmenting the cell cluster area image, thereby ensuring the accuracy of subsequent cell cluster area images, and further improving the recognition accuracy of the target type of the target cell in the digital slice image to be detected.

[0084] S103. For each of the single cell region images in each of the cell cluster region images, the single cell region image is compared with the standard cell image to determine whether the single cell in the single cell region image is a suspicious cell.

[0085] In this step, each of the single cell area images in each of the cell cluster area images is compared with the standard cell image, the distance between each of the single cell area images and the standard cell image is determined, and the distance is compared with a preset distance to determine whether the single cell in the single cell area image is a suspicious cell.

[0086] In this way, when the distance is less than the distance, the single cell in the single cell area image is determined to be a suspicious cell.

[0087] Here, the size of the preset distance can be customized according to different application scenarios, actual needs and actually required suspicious cell accuracy, and there is one and only one standard cell image in each of the cell cluster area images.

[0088] S104, inputting the suspicious cell images corresponding to the suspicious cells into the trained cell classification model to determine whether there are target cells of the target type in the digital slice image.

[0089] In this step, the target cells are cells of the target type that the staff wants to identify from the digital slice image to be tested. For example, the embodiment provided in the present application takes the digital slice image to be tested of identifying cervical epithelial cells as an example, and wants to determine the target cells with positive target type and the negative cells with negative target type among the atrophic and clustered cells, so as to help the staff as an auxiliary medical reference.

[0090] In this way, the suspicious cell images corresponding to each suspicious cell are input into the trained cell classification model, and the trained cell classification model outputs the confidence value (between 0 and 1.0) of the target cell corresponding to each suspicious cell being a positive target type, and then the suspicious cells whose confidence values ​​exceed the preset confidence values ​​are determined as target cells.

[0091] In the above, the trained cell classification model can adopt but is not limited to the network classification models of the ResNet series, Inception series and SeNet. The embodiment provided in this application takes the Resnet-101 type network classification model as an example.

[0092] Here, the training data set and validation data set of the trained cell classification model are determined in the following way:

[0093] The sample single cell area images are labeled with target cells according to the target type, and the labeled target types are set as training data sets and validation data sets according to a certain preset ratio.

[0094] In the above, the labeled target type is used for target cells, different initial cell classification models using different network parameters are trained, and then the cell classification model with the most accurate training result is selected as the trained cell classification model in the embodiment of the present application.

[0095] The method for identifying cells in cell clusters provided in the embodiments of the present application is different from the cell identification method in the prior art in that the embodiment provided in the present application compares the single cell area image in the digital slice image to be detected with the standard cell image, and further accurately identifies the suspicious cells after the comparison, thereby determining the target type of the target cells in the digital slice image, thereby realizing the use of a dual detection and identification model, and improving the accuracy of identifying target cells in normal cell clusters and atrophic cell clusters.

[0096] See also Figure 3 , Figure 3 This is a second flow chart of a method for identifying cells in a cell cluster provided in another embodiment of the present application. Figure 3 As shown in , the method for identifying cells in a cell cluster provided in the embodiment of the present application includes:

[0097] S301 . Identify at least one cell cluster region image from the digital slice image to be detected.

[0098] S302 . For each of the cell cluster region images, identify individual single cell region images and a standard cell image from the cell cluster region image.

[0099] S303. For each of the single cell region images in each of the cell cluster region images, the single cell region image and the standard cell image are input into a trained cell screening model to determine the similarity between each single cell in the cell cluster region image and the standard cell.

[0100] In this step, for each single cell region image in the cell cluster region image, the single cell region image and the standard cell image are input into the trained cell screening model, the image features of each single cell region image and the image features of the standard cell image are extracted, and the similarity between each single cell in the cell cluster region image and the standard cell is determined by the cosine distance or Euclidean distance between the image features of the single cell region image and the image features of the standard cell image. That is, the trained cell screening model outputs the cosine distance or Euclidean distance between the trained cell screening model and the standard cell image.

[0101] Among them, the trained cell screening model can adopt but is not limited to a ResNet-51 type network screening model.

[0102] Optionally, confirm the trained cell screening model by:

[0103] Obtain different types of sample single cell area images in the sample digital slice image, labels of each of the sample single cell area images, and sample standard cell images corresponding to the sample digital slice image; the label of the sample single cell area image is used to characterize the position information and area size of the sample single cell area image.

[0104] The sample single cell area image and the label of the sample single cell area image are input into the initial cell screening model to determine the preset sample distances between different types of sample single cells and sample standard cells.

[0105] The preset sample distance between different types of sample single cells and sample standard cells may specifically be a cosine distance between the sample single cells and the sample standard cells.

[0106] When the loss value between the preset sample distance and the actual sample distance between different types of sample single cells and the sample standard cells is less than a preset threshold, the training is terminated and the trained cell screening model is determined.

[0107] The loss value between the preset sample similarity and the real sample similarity is determined by the following formula:

[0108] Y=|A|,A<0;

[0109] Y=0,A>0;

[0110] in,

[0111] Wherein, Y is used to represent the loss value between the preset sample similarity and the real sample similarity, i is used to represent the number of feature vectors of the sample single cell, and j is used to represent the number of feature vectors of the sample standard cell; v bj The feature vector used to characterize the sample standard cell; v ai and v ci The feature vectors used to characterize the first type of sample single cells and the feature vectors used to characterize the second type of sample single cells respectively; N is used to characterize the number of sample single cells of the first type; 3N is the number of sample single cells of the second type.

[0112] Here, the “10” in the formula is used to represent that the number of standard cells in the sample is 10.

[0113] In the embodiments provided in the present application, N>=32.

[0114] In the above, the distances between different types of sample single-cell area images are different. This application takes the positive cells and negative cells of the atrophic cell clusters in the digital section image of cervical epithelial cells as an example. The distance between the positive cell area image and the sample single-cell area image is greater than the distance between the negative cells and the sample single-cell area image.

[0115] S304: If the similarity is less than a preset threshold, determining the single cell corresponding to the similarity as a suspicious cell in the cell cluster region image.

[0116] In this step, if the similarity between different types of sample single cells and sample standard cells is less than the preset threshold, that is, the cosine distance or Euclidean distance between the cell screening model and the standard cell image is less than the preset threshold, it means that the probability that this type of sample single cell is a suspicious cell is relatively high.

[0117] Among them, the preset threshold can be customized according to the actual application scenario.

[0118] S305 , inputting the suspicious cell images corresponding to the suspicious cells into the trained cell classification model to determine whether there are target cells of the target type in the digital slice image.

[0119] The description of S301 to S302 and S305 may refer to the description of S101 to S102 and S104, and can achieve the same technical effect, which will not be elaborated herein.

[0120] The method for identifying cells in cell clusters provided in the embodiments of the present application is different from the cell identification method in the prior art in that the embodiment provided in the present application compares the single cell area image in the digital slice image to be detected with the standard cell image, and further accurately identifies the suspicious cells after the comparison, thereby determining the target type of the target cells in the digital slice image, thereby realizing the use of a dual detection and identification model, and improving the accuracy of identifying target cells in normal cell clusters and atrophic cell clusters.

[0121] The following is a specific example to illustrate the process of the method for identifying cells in a cell cluster provided in the embodiment of the present application. Figure 4 As shown, Figure 4 A flow chart of an embodiment of a method for identifying cells in a cell cluster is shown. The specific process is as follows:

[0122] S401. Obtain a digital slice of cervical cytology to be tested.

[0123] The scanning magnification of the digital slice to be detected is 40 times, and the size of the scanned section is 20000x20000.

[0124] S402 , decoding the digital slice to be detected according to a corresponding decoding program to obtain a digital slice image to be detected, and determining a standard cell image of the digital slice image to be detected.

[0125] Among them, the standard cell image can be used to represent I b To express.

[0126] S403 , scaling the digital slice image to be detected to obtain two scaled digital slice images to be detected.

[0127] Among them, when the zoom factors are 20 times and 10 times, the sizes of the two digital slice images to be detected after zooming are 10000x10000 (corresponding to 20 times) and 5000x5000 (corresponding to 10 times).

[0128] S404 , performing image block processing on the digital slice images to be detected at the different resolutions to obtain image blocks of preset sizes.

[0129] The size of each image block is 512x512, and each image block overlaps by 16 pixels in the horizontal and vertical directions.

[0130] S405. For each image block, a trained fine cell cluster detection model corresponding to the current resolution is selected from three corresponding trained cell cluster detection models to detect cell clusters, and the position information and size relationship of the cell clusters in the digital slice images to be detected at the three resolutions are obtained respectively, and the non-maximum suppression method is used to merge them to obtain each single cell area image.

[0131] After obtaining the position information and size relationship of the cell clusters, the position information and size relationship of the cell clusters need to be mapped back to the current maximum multiple resolution, and each single cell region image can be mapped using {I i , i=1,...,M}.

[0132] S406, determining whether the target distance between any two single-cell region images is greater than or equal to a preset threshold; if so, determining the target region image between the any two single-cell region images in the cell cluster region image as a single-cell region image.

[0133] Among them, each single cell area image is represented as I i，j , j=1,...,R, and the preset threshold can be designed as 1.5 times the single cell area image according to the actual application scenario.

[0134] S407, determine each single cell region image I i，j Compared with the standard cell image I b The distance between them is used to select two single cell region images I whose cosine distance is less than the preset distance. i，j As suspicious positive cells.

[0135] Among them, each single cell area image I i，j Compared with the standard cell image I b The distance between them can be cosine distance or Euclidean distance.

[0136] S408. Input the above-mentioned suspicious positive cells into the trained cell classification model to determine whether there are target cells of the target type in the digital slice image. If so, determine the confidence value of the target cell. According to the confidence value of the target cell, determine the target type of the target cell, and feed back the recognition result to the staff.

[0137] Among them, the staff can judge whether there are target positive cells or target negative cells in the digital slice image according to the above feedback results.

[0138] See also Figure 5 , Figure 5 A schematic diagram of a structure of a device for identifying cells in a cell cluster provided in an embodiment of the present application, wherein the device 500 for identifying cells in a cell cluster comprises:

[0139] The first identification module 510 is used to identify at least one cell cluster region image from the digital slice image to be detected.

[0140] The second identification module 520 is used to identify, for each of the cell cluster region images, individual single cell region images and a standard cell image from the cell cluster region image.

[0141] Optionally, the second identification module 520 is specifically configured to:

[0142] For each of the cell cluster region images, each cell nucleus is identified from the cell cluster region image, and the cell image corresponding to each cell nucleus is determined as a single cell region image.

[0143] It is determined whether the target distance between any two single cell region images in the cell cluster region image is greater than or equal to a preset threshold.

[0144] If it is greater than, the target area image between any two single cell area images in the cell cluster area image is determined as a single cell area image.

[0145] The first determination module 550 is used to compare each of the single cell region images in each of the cell cluster region images with the standard cell image to determine whether the single cell in the single cell region image is a suspicious cell.

[0146] Optionally, the first determining module 550 is specifically configured to:

[0147] For each of the single cell region images in each of the cell cluster region images, the single cell region image and the standard cell image are input into a trained cell screening model to determine the similarity between each single cell in the cell cluster region image and the standard cell.

[0148] The trained cell screening model is obtained by training different types of sample single cell area images in the sample digital slice images and sample standard cell images in the sample digital slice images.

[0149] If the similarity is less than a preset threshold, the single cell corresponding to the similarity is determined as a suspicious cell in the cell cluster region image.

[0150] Optionally, confirm the trained cell screening model by:

[0151] Obtain different types of sample single cell area images in the sample digital slice image, type labels of each of the sample single cell area images, and sample standard cell images corresponding to the sample digital slice image; the type label is used to characterize the real sample similarity between the sample single cell and the preset sample standard cell.

[0152] The sample single cell region image and the label of the sample single cell region image are input into an initial cell screening model to determine a preset sample similarity between the sample single cell and a sample standard cell.

[0153] When the loss value between the preset sample similarity and the real sample similarity between the sample single cell and the sample standard cell is less than a preset threshold, the training is terminated and the trained cell screening model is determined.

[0154] Optionally, the loss value between the preset sample similarity and the real sample similarity is determined by the following formula:

[0155] Y=|A|,A<0;

[0156] Y=0,A>0;

[0157] in,

[0158] Wherein, Y is used to represent the loss value between the preset sample similarity and the real sample similarity, i is used to represent the number of feature vectors of the sample single cell, and j is used to represent the number of feature vectors of the sample standard cell; v bj The feature vector used to characterize the sample standard cell; v ai and v ci The feature vectors used to characterize the first type of sample single cells and the feature vectors used to characterize the second type of sample single cells respectively; N is used to characterize the number of sample single cells of the first type; 3N is the number of sample single cells of the second type.

[0159] The second determination module 540 is used to input the suspicious cell images corresponding to the suspicious cells into the trained cell classification model to determine whether there are target cells of the target type in the digital slice image.

[0160] In the above, the application scenarios of the identification device 500 for cells in a cell cluster provided in the embodiment of the present application depend on the types of different cell clusters, and here, include but are not limited to the following two application scenarios:

[0161] Application scenario 1:

[0162] The identification device 500 for cells in a cell cluster provided in the present application can be directly used to identify the target type of the target cell in the digital slice image to be detected, and to provide the identification result to the staff for further judgment.

[0163] Application scenario 2:

[0164] The identification device 500 for cells in cell clusters provided in the present application can be used in combination with existing cytology auxiliary diagnosis products or systems, and the identification results can be used as auxiliary data, combined with the detection results of existing cytology auxiliary products, and provided to the staff for reference.

[0165] The identification device 500 for cells in cell clusters provided in the embodiment of the present application, compared with the cell identification method in the prior art, compares the single-cell area image in the digital slice image to be detected with the standard cell image, and further accurately identifies the suspicious cells after the comparison, determines the target type of the target cell in the digital slice image, and realizes the use of a dual detection and identification model, thereby improving the accuracy of identifying target cells in normal cell clusters and atrophic cell clusters.

[0166] See also Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 6 As shown in , the electronic device 600 includes a processor 610 , a memory 620 and a bus 630 .

[0167] The memory 620 stores machine-readable instructions executable by the processor 610. When the electronic device 600 is running, the processor 610 communicates with the memory 620 via the bus 630. When the machine-readable instructions are executed by the processor 610, the above-mentioned Figure 1 as well as Figure 2 The steps of the method for identifying cells in a cell cluster in the method embodiment shown are specifically implemented in the method embodiment and will not be described in detail here.

[0168] The present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the computer program can execute the above-mentioned Figure 1 as well as Figure 2 The steps of the method for identifying cells in a cell cluster in the method embodiment shown are specifically implemented in the method embodiment and will not be described in detail here.

[0169] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0170] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple 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 communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0171] 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 multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0172] 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.

[0173] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application can be embodied in the form of a software product in essence or in other words, the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0174] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes 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. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for identifying cells in a cell cluster, It is characterized in that The method for identifying cells in a cell cluster comprises: identifying at least one cell cluster region image from the digital slice image to be detected; For each of the cell cluster region images, identifying individual single cell region images and a standard cell image from the cell cluster region image; For each of the single cell region images in each of the cell cluster region images, the single cell region image is compared with the standard cell image to determine whether the single cell in the single cell region image is a suspicious cell; Inputting the suspicious cell images corresponding to the suspicious cells into the trained cell classification model to determine whether there are target cells of the target type in the digital slice image; For each of the single cell region images in each of the cell cluster region images, the single cell region image is compared with the standard cell image to determine whether the single cell in the single cell region image is a suspicious cell, including: For each of the single cell region images in each of the cell cluster region images, the single cell region image and the standard cell image are input into a trained cell screening model to determine the similarity between each single cell in the cell cluster region image and the standard cell; If the similarity is less than a preset threshold, determining the single cell corresponding to the similarity as a suspicious cell in the cell cluster region image; Determine the trained cell screening model by: Obtaining different types of sample single cell area images in the sample digital slice image, type labels of each of the sample single cell area images, and sample standard cell images corresponding to the sample digital slice image; the type label is used to characterize the real sample similarity between the sample single cell and the preset sample standard cell; Inputting the sample single cell region image and the label of the sample single cell region image into an initial cell screening model to determine a preset sample similarity between the sample single cell and a sample standard cell; When the loss value between the preset sample similarity and the real sample similarity between the sample single cell and the sample standard cell is less than a preset threshold, the training is terminated and the trained cell screening model is determined; The loss value between the preset sample similarity and the real sample similarity is determined by the following formula: ，A<0； ，A＞0； in, ; in, It is used to characterize the loss value between the preset sample similarity and the real sample similarity, The number of feature vectors used to characterize the single cells of the sample, The number of feature vectors used to characterize the sample standard cells; Feature vector used to characterize the sample standard cells; and A feature vector for characterizing a first type of sample single cell and a feature vector for characterizing a second type of sample single cell; Used to characterize the number of single cells in the first type of sample; The number of single cells in the second type of sample.

2. The method for identifying cells in a cell cluster according to claim 1, It is characterized in that For each of the cell cluster region images, identifying each single cell region image from the cell cluster region image includes: For each of the cell cluster region images, identifying each cell nucleus from the cell cluster region image, and determining the cell image corresponding to each cell nucleus as a single cell region image; Determine whether the target distance between any two single cell region images in the cell cluster region image is greater than or equal to a preset threshold; If it is greater than, the target area image between any two of the single cell area images in the cell cluster area image is determined as the single cell area image.

3. A device for identifying cells in a cell cluster, It is characterized in that The identification device suitable for cells in a cell cluster comprises: A first recognition module is used to recognize at least one cell cluster region image from the digital slice image to be detected; A second recognition module is used to recognize, for each of the cell cluster region images, individual single cell region images and a standard cell image from the cell cluster region image; A first determination module is used to compare each of the single cell region images in each of the cell cluster region images with the standard cell image to determine whether the single cell in the single cell region image is a suspicious cell; A second determination module is used to input the suspicious cell images corresponding to the suspicious cells into the trained cell classification model to determine whether there are target cells of the target type in the digital slice image; The first determining module is specifically configured to: For each of the single cell region images in each of the cell cluster region images, the single cell region image and the standard cell image are input into a trained cell screening model to determine the similarity between each single cell in the cell cluster region image and the standard cell; If the similarity is less than a preset threshold, determining the single cell corresponding to the similarity as a suspicious cell in the cell cluster region image; Determine the trained cell screening model by: Obtaining different types of sample single cell area images in the sample digital slice image, type labels of each of the sample single cell area images, and sample standard cell images corresponding to the sample digital slice image; the type label is used to characterize the real sample similarity between the sample single cell and the preset sample standard cell; Inputting the sample single cell region image and the label of the sample single cell region image into an initial cell screening model to determine a preset sample similarity between the sample single cell and a sample standard cell; When the loss value between the preset sample similarity and the real sample similarity between the sample single cell and the sample standard cell is less than a preset threshold, the training is terminated and the trained cell screening model is determined; The loss value between the preset sample similarity and the real sample similarity is determined by the following formula: ，A<0； ，A＞0； in, ; in, It is used to characterize the loss value between the preset sample similarity and the real sample similarity, The number of feature vectors used to characterize the single cells of the sample, The number of feature vectors used to characterize the sample standard cells; Feature vector used to characterize the sample standard cells; and A feature vector for characterizing a first type of sample single cell and a feature vector for characterizing a second type of sample single cell; Used to characterize the number of single cells in the first type of sample; The number of single cells in the second type of sample.

4. The device for identifying cells in a cell cluster according to claim 3, It is characterized in that The second identification module is specifically used for: For each of the cell cluster region images, identifying each cell nucleus from the cell cluster region image, and determining the cell image corresponding to each cell nucleus as a single cell region image; Determine whether the target distance between any two single cell region images in the cell cluster region image is greater than or equal to a preset threshold; If it is greater than, the target area image between any two of the single cell area images in the cell cluster area image is determined as the single cell area image.

5. An electronic device, It is characterized in that include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate via the bus, and the machine-readable instructions are executed by the processor to perform the steps of the method for identifying cells in a cell cluster as described in any one of claims 1 to 2.

6. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for identifying cells in a cell cluster as described in any one of claims 1 to 2 are executed.

Images