Cell information statistics method, device, equipment and computer readable storage medium

Abstract

The present invention discloses a method, device, equipment and computer-readable storage medium for counting cell information, and the method includes: S1, obtaining target image information of a cell group to be counted; S2, selecting a pre-built model, and inputting the target image information into the pre-built model to obtain a result image; S3, judging whether there is a defect feature in a region of the result image, the defect feature includes cell adhesion, at least partial disappearance of cells, and several items of cell abnormality, if so, executing steps S4-S6, otherwise executing step S6; S4, marking at least one region with defect features; S5, retraining the pre-built model according to the marking operation, and inputting the target image information into the retrained model to obtain a new result image; S6, counting cells based on the current result image. The present invention adopts deep learning combined with a semi-supervised learning method, and only a small amount of marking is required for the region with unsatisfactory results to achieve the desired effect.

Description

Technical Field

[0001] The present application relates to the field of cell image analysis, and in particular to a cell information statistical method, device, equipment and computer-readable storage medium. Background Art

[0002] Cell counting is a hot topic in the field of computer vision. Extracting and counting cell information is an important technology and a difficult problem in medical image processing. Currently, cells are usually segmented and counted based on traditional deep learning methods. Traditional deep learning methods include building a data set, data amplification, designing a deep learning model, and then training on the data set until the model converges to obtain a basic model.

[0003] Such a deep learning method requires the collection of a large number of training images during the data set construction stage, and they need to be labeled one by one. If the obtained model needs to recognize cells of other cell lines, a large number of training images of other cell lines need to be collected again and labeled one by one.

[0004] In order to make up for the lack of accuracy of current deep learning models, fluorescent cell images or stained cell images are usually used to enhance cell features. However, fluorescence and staining not only increase costs, but are also toxic to cells. In addition, it takes a lot of time to pre-fluorescently excite or stain cells. For example, DAPI staining takes about 70 minutes, and the subsequent image processing steps also become cumbersome.

[0005] Currently, the traditional random forest machine learning algorithm is commonly used to count non-stained or non-fluorescent cells, but this algorithm is not accurate enough and requires users to perform a lot of annotations to obtain ideal results. Summary of the invention

[0006] The purpose of the present invention is to provide an image processing algorithm that can realize cell segmentation and counting with only a small amount of annotation, so as to realize the statistics of cell information.

[0007] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0008] A cell information statistics method is used to perform information statistics on a cell group, comprising the following steps:

[0009] S1, obtaining target image information of a cell group to be counted;

[0010] S2, selecting a pre-built model, and inputting the target image information into the pre-built model to obtain a result image;

[0011] S3, determining whether there are defective features in any area of ​​the result image, wherein the defective features include one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality; if so, executing steps S4-S6; otherwise, executing step S6;

[0012] S4, marking at least one area having defective features;

[0013] S5, retraining the pre-built model according to the labeling operation, and inputting the target image information into the retrained model to obtain a new result image;

[0014] S6. Based on the current result image, count the cell information, where the cell information includes one or more of the number of cells, the size of cells, and the shape of cells.

[0015] Further, after step S5, steps S3 to S5 are repeatedly performed until a preset number of repetitions is reached, or until there is no area with defective features in the result image, and then step S6 is performed.

[0016] Furthermore, the pre-built model is trained by the following steps:

[0017] M1. Obtain multiple images of the cell line and manually annotate them to obtain annotated learning samples;

[0018] M2. Design a deep learning model, and use the labeled learning samples to train the deep learning model to obtain a basic model;

[0019] M3. Retrain the basic model using a semi-supervised learning method, wherein the semi-supervised learning method obtains the pre-built model by focusing on and learning only the areas where the annotation operation is performed.

[0020] Furthermore, step M3 further includes:

[0021] M301. Obtain a sample image set for establishing a pre-built model, wherein the sample image set includes a plurality of sample images;

[0022] M302, input one of the sample images into the basic model to obtain a training result image;

[0023] M303, determining an area in the training result image where defect features exist, where the defect features include one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality;

[0024] M304. Mark at least one area with defective features;

[0025] M305, retraining the current model according to the labeling operation, focusing on and learning only the area where the labeling operation is performed during the retraining process, and obtaining a new training result image based on the retrained current model;

[0026] Repeat steps M303 to M305 until a preset number of repetitions is reached, or until there is no area with defective features in the training result image, and then execute step M306;

[0027] M306, updating the basic model to the current model;

[0028] M307, traverse the sample image set, input the next sample image into the current model to obtain the result image in the current training, and execute steps M303 to M306 until the sample image has been traversed for a preset number of rounds, and then execute M308;

[0029] M308. Save the current model to obtain the pre-built model.

[0030] Furthermore, in the process of retraining the pre-built model according to the labeling operation in step S5, only the area where the labeling operation is performed is focused on and learned, including: extracting features according to the labeling operation to retrain the model.

[0031] Furthermore, the focusing on and learning only the area where the annotation operation is performed includes:

[0032] Delimiting the region for the labeling operation so that the defined region includes the cells involved in the labeling operation;

[0033] Generate a sub-image of the defined area, and input the sub-image and corresponding annotation operation information into the model to be retrained as a deep learning sample;

[0034] The corresponding annotation operation information is used as a learning target, and the features of the sub-image are deeply learned to update the neural network parameters in the model to obtain a retrained model.

[0035] Furthermore, after step M3, the method further includes:

[0036] M4. Obtain multiple images of other cell lines and manually annotate them to obtain annotated samples of other cell lines;

[0037] M5. Input the labeled samples of other cell lines into the pre-built model, and retrain it using a semi-supervised learning method to obtain a pre-built model that supports the other cell lines.

[0038] Furthermore, the labeling operation includes background labeling and / or cell labeling in the form of manual labeling.

[0039] Further, in step S2, one of the pre-built models is selected from the pre-built model database, and before or after step S6, the following steps are further included:

[0040] S7. Save the current retrained model and add it to the model database.

[0041] Optionally, the target image information of the cell population to be counted in step S1 is one or more of a non-stained and non-fluorescent cell image, a stained cell image, a fluorescent cell image, a phase contrast image, and a bright field image.

[0042] According to another aspect of the present invention, a cell information statistics device is provided for performing information statistics on cells in a cell group, comprising the following modules:

[0043] A target acquisition module to be counted, which is configured to acquire target image information of a cell group to be counted;

[0044] A pre-built model calling module is configured to select a pre-built model and input the target image information into the pre-built model to obtain a result image;

[0045] A defect feature determination module, configured to determine an area in the result image where defect features exist, wherein the defect features include one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality;

[0046] a labeling and retraining module, configured to receive the determination result of the defect feature determination module, label at least one area where the defect feature exists, retrain the pre-built model according to the labeling operation, and input the target image information into the retrained model to obtain a new result image;

[0047] The statistical module is configured to perform statistics on cell information based on the result image, wherein the cell information includes one or more of cell number, cell size, and cell shape.

[0048] Furthermore, the pre-built model calling module includes the following units:

[0049] A labeled sample unit, which is configured to obtain multiple images of the cell line and manually label them to obtain labeled learning samples;

[0050] A basic model unit, configured to design a deep learning model and train the deep learning model using the labeled learning samples to obtain a basic model;

[0051] The pre-built model unit is configured to retrain the basic model using a semi-supervised learning method, wherein the semi-supervised learning method obtains the pre-built model by focusing on and learning only the area where the annotation operation is performed.

[0052] Furthermore, the pre-built model unit includes the following sub-units:

[0053] A sample image set acquisition subunit, configured to acquire a sample image set for establishing a pre-built model, wherein the sample image set includes a plurality of sample images;

[0054] A traversal subunit, configured to traverse the sample image set and send one of the sample images to the training subunit each time;

[0055] A training subunit, configured to input the received sample image into the basic model to obtain a training result image;

[0056] a defect determination subunit, configured to receive the training result image obtained by the training subunit, and determine whether there is a defect feature in a region of the training result image, wherein the defect feature includes one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality;

[0057] A marking and retraining subunit is configured to mark at least one area with defect features; then retrain the basic model according to the marking operation, only focusing on and learning the area where the marking operation is performed during the retraining process, and obtaining a new training result image based on the retrained current model, and sending it to the defect identification subunit;

[0058] A model updating subunit, configured to update the basic model to the current model when the defect identification subunit identifies a defect-free feature; and trigger the traversal subunit to send the next sample image to the training subunit;

[0059] The model saving subunit is configured to save the current model in response to the completion of the traversal process of the traversal subunit to obtain the pre-built model.

[0060] According to another aspect of the present invention, an electronic device is provided, comprising a processor and a memory, wherein the memory is used to store program instructions, and the processor is configured to run the program instructions, and the program instructions are run to perform the steps of the method described above.

[0061] According to another aspect of the present invention, a computer-readable storage medium is provided for storing program instructions, wherein the program instructions are configured to be called to execute the steps of the method described above.

[0062] According to another aspect of the present invention, a computer program product is provided, comprising a readable and stored computer program, wherein the computer program comprises program instructions, and when the program instructions are executed on a computer device, the computer device executes the steps of the above-mentioned method.

[0063] According to another aspect of the present invention, a model training method is provided, comprising the following steps:

[0064] M1. Obtain multiple images of the cell line and manually annotate them to obtain annotated learning samples;

[0065] M2. Design a deep learning model, and use the labeled learning samples to train the deep learning model to obtain a basic model;

[0066] M3. Retrain the basic model using a semi-supervised learning method, wherein the semi-supervised learning method obtains the pre-built model by focusing on and learning only the region where the annotation operation is performed;

[0067] Wherein, step M3 further comprises:

[0068] M301. Obtain a sample image set for establishing a pre-built model, wherein the sample image set includes a plurality of sample images;

[0069] M302, input one of the sample images into the basic model to obtain a training result image;

[0070] M303, determining an area in the training result image where defect features exist, where the defect features include one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality;

[0071] M304. Mark at least one area with defective features;

[0072] M305, retraining the current model according to the labeling operation, focusing on and learning only the area where the labeling operation is performed during the retraining process, and obtaining a new training result image based on the retrained current model;

[0073] Repeat steps M303 to M305 until a preset number of repetitions is reached, or until there is no area with defective features in the training result image, and then execute step M306;

[0074] M306, updating the basic model to the current model;

[0075] M307, traverse the sample image set, input the next sample image into the current model to obtain the result image in the current training, and execute steps M303 to M306 until the sample image has been traversed for a preset number of rounds, and then execute M308;

[0076] M308. Save the current model to obtain the pre-built model.

[0077] The beneficial effects brought by the technical solution provided by the present invention are as follows:

[0078] a. If the user is not satisfied with the output of the pre-built model, the user can annotate the result image and retrain the model to achieve the desired effect. Therefore, the accuracy requirement for the initial model is relatively low, and a large number of annotated training samples are not required in the early stage of model building;

[0079] b. The basic model obtained through deep learning is combined with a small number of annotated cell line images, and a pre-built model of the cell line can be obtained through semi-supervised learning. The pre-built models of different cell lines can share the same basic model of deep learning, that is, the same basic model combined with different cell line images can obtain pre-built models corresponding to different cell lines;

[0080] c. Applicable to most cell images, including stained cell images, fluorescent cell images, phase contrast images, bright field images, and can even accurately distinguish cells in non-stained and non-fluorescent cell images;

[0081] d. The semi-supervised learning method only focuses on and learns the areas where annotation operations are performed. Compared with the fully supervised learning method, it has the advantages of simpler algorithms and clearer retraining learning objectives. Users can annotate the image defect areas that they are not satisfied with and retrain the model to obtain a satisfactory model. As users identify more and more cell images, the model accuracy becomes higher and higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the conventional technical solutions, the drawings required for use in the description of the embodiments or the conventional technical solutions will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0083] Figure 1 It is a basic flow chart of a cell information statistics method provided by an exemplary embodiment of the present invention;

[0084] Figure 2 It is a specific schematic diagram of a cell information statistics method provided by an exemplary embodiment of the present invention;

[0085] Figure 3 is a flowchart of a training method for a pre-built model provided by an exemplary embodiment of the present invention;

[0086] Figure 4 is a flowchart of a method for retraining the basic model using a semi-supervised learning method provided by an exemplary embodiment of the present invention;

[0087] Figure 5 is a schematic flow chart of a method for establishing a pre-built model supporting more than two cell lines provided by an exemplary embodiment of the present invention;

[0088] Figure 6 is a flow chart of a method for focusing on and learning only the area where the annotation operation is performed, provided by an exemplary embodiment of the present invention;

[0089] Figure 7 is a schematic block diagram of modules of a cell information statistics device provided by an exemplary embodiment of the present invention;

[0090] Figure 8 is a schematic block diagram of subunits of a pre-built model unit provided by an exemplary embodiment of the present invention;

[0091] Figure 9(a) is the original image of the non-stained and non-fluorescent cell image;

[0092] FIG9(b) is a result image obtained after inputting the pre-built model into FIG9(a) provided by an exemplary embodiment of the present invention; FIG9(c) is an image after marking the defect features in FIG9(b) provided by an exemplary embodiment of the present invention;

[0093] Figure 10(a) is the original image of the fluorescent cell image;

[0094] FIG. 10( b ) is a result image obtained after FIG. 10( a ) is input into a pre-built model according to an exemplary embodiment of the present invention;

[0095] FIG. 10( c ) is an image after the defect features in FIG. 10( b ) are annotated, provided by an exemplary embodiment of the present invention;

[0096] FIG. 10( d ) is a result image output after retraining the model using FIG. 10( c ) according to an exemplary embodiment of the present invention;

[0097] Fig.11 is a schematic diagram of a system 100 for executing the method described in an embodiment of the present invention. DETAILED DESCRIPTION

[0098] In order to enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0099] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0100] In conventional methods of counting cells using image analysis, cells need to be stained or fluorescently excited so that they present specific features in the image in order to distinguish adjacent cells. The purpose of the present invention is to provide an improved image analysis method that can accurately count the cell information in the image through simple operations, even for unstained and non-fluorescent cell images. Since unstained and non-fluorescent cell images are the most difficult type of images to analyze and process, the cell information counting method applicable to unstained and non-fluorescent cell images in the present invention is also applicable to other types such as stained cell images, fluorescent cell images, phase contrast images, and bright field images.

[0101] The present invention uses image data of a cell line for deep learning and trains a basic model; then based on this basic model, a semi-supervised learning method is used to train a pre-built model, and in response to support needs of other cell lines, only a small number of labeled images of other cell lines are needed as training samples for semi-supervised learning, without the need to re-establish the basic model.

[0102] In one embodiment of the present invention, a cell information statistics method is provided for performing information statistics on a cell group, such as counting the number of cells, classifying and measuring the size of cells, or classifying and measuring the shape of cells.

[0103] In this embodiment, one or more pre-built models need to be established in advance to form a model database, and a suitable pre-built model is selected according to the cell population to be statistically analyzed, and the pre-built model is used to process the cell image to obtain the result image. Figure 3 As shown, the pre-built models in the database are trained through the following steps:

[0104] M1. Obtain multiple images of a cell line and manually annotate them to obtain annotated learning samples;

[0105] M2. Design a deep learning model, and use the labeled learning samples to train the deep learning model to obtain a basic model;

[0106] M3. Retrain the basic model using a semi-supervised learning method, wherein the semi-supervised learning method obtains the pre-built model by focusing on and learning only the areas where the annotation operation is performed.

[0107] Specific as Figure 4 As shown, step M3 further includes:

[0108] M301. Obtain a sample image set for establishing a pre-built model, wherein the sample image set includes multiple sample images.

[0109] Different from the labeled learning samples in step M1, the sample images in the sample image set here do not need to be manually labeled in advance. For example, there are 100 sample images in the sample image set, which are numbered from 0001 to 0100.

[0110] M302, input one of the sample images into the basic model to obtain a training result image;

[0111] M303, determining an area in the training result image where defect features exist, where the defect features include one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality;

[0112] M304. Mark at least one area with defective features;

[0113] M305, retraining the current model according to the labeling operation, specifically, inputting the sample image and at least the local area image of the labeling operation (or the training result image after the labeling operation) into the current model to retrain and obtain a new model, during the retraining process, only focusing on and learning the area where the labeling operation is performed, and obtaining a new training result image based on the retrained current model;

[0114] Repeat steps M303 to M305 until a preset number of repetitions is reached, or until there is no area with defective features in the training result image, and then execute step M306;

[0115] M306, updating the basic model to the current model;

[0116] For example, the basic model is retrained using a semi-supervised learning method using a single sample image in the order of numbers 0001 to 0100, and the training of the sample image numbered 0001 is now completed.

[0117] M307, traverse the sample image set, input the next sample image into the current model to obtain the result image in the current training, and execute steps M303 to M306 until the sample image has been traversed for a preset number of rounds, and then execute M308;

[0118] For example, the sample image numbered 0002 is traversed in order, and steps M302 to M306 are executed on it, and then steps M302 to M306 are executed on the sample image numbered 0003, until the sample image numbered 0100 is traversed, and steps M302 to M306 are executed on it. So far, one round of traversal is completed, and the present invention does not limit the number of rounds of traversal. If all images in the sample image set are not completely traversed (for example, the image numbered 0066 is skipped, or only the image numbered 0098 is traversed), it should also be regarded as a simple replacement of this embodiment, and also falls within the protection scope required by the present invention.

[0119] M308. Save the current model to obtain the pre-built model.

[0120] After completing the traversal according to the preset traversal rules, the final model obtained is the pre-built model.

[0121] In one embodiment of the present invention, a pre-built model supporting more than two cell lines can also be established, such as Figure 5 As shown:

[0122] In step M1, assuming that multiple images of the first cell line are acquired, then the corresponding step M308 obtains a pre-built model supporting the first cell line;

[0123] After step M3, the method further includes:

[0124] M4. Obtain multiple images of other cell lines (such as the second cell line), and manually annotate them to obtain annotated samples of other cell lines;

[0125] M5. Input the labeled samples of other cell lines into the pre-built model, and retrain it using a semi-supervised learning method to obtain a pre-built model that supports the other cell lines.

[0126] The specific execution steps of step M5 are as follows: using a sample image set that is the same as or different from step M301, inputting the first sample image into the pre-built model to obtain a training result image;

[0127] Determine an area in the training result image where defect features exist, wherein the defect features include one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality; perform a labeling operation on at least one area where defect features exist; retrain the basic model according to the labeling operation, and during the retraining process, only pay attention to and learn the area where the labeling operation is performed, and obtain a new training result image based on the retrained current model; repeat this step until a preset number of repetitions is reached, or until there is no area with defect features in the training result image, and then perform the following steps:

[0128] The basic model is updated to the current model, the next sample image in the sample image set is traversed, and it is input into the current model to obtain the result image currently being trained, and the steps in the previous paragraph are executed, that is, each time a sample image is traversed, the current model is updated and saved once, until the sample images in the sample image set have undergone a preset round of traversal, and the latest current model is saved as a pre-built model that supports both the first cell line and the second cell line.

[0129] If it is desired to support the third cell line at the same time, multiple images of the second cell line may be obtained, and steps M4 to M5 may be performed, which will not be described in detail here.

[0130] Steps M1 to M3 may also be performed for different cell lines to obtain different pre-built models, and these pre-built models may be saved to obtain a pre-built model database.

[0131] Based on the establishment of pre-built models, cell images (even non-stained and non-fluorescent cell images) can be identified to count the cell information in the image (such as number, size or shape), such as Figure 1 As shown, the cell information statistical method includes the following steps:

[0132] S1. Obtain target image information of the cell population to be counted.

[0133] The cell group to be counted is often not a single cell, but a state of multiple cells aggregated. The cell group can be imaged using a microscope to obtain a target image, such as the non-stained and non-fluorescent cell image shown in FIG9(a).

[0134] S2. Select a pre-built model and input the target image information into the pre-built model to obtain a result image, as shown in FIG9( b ).

[0135] Specifically, one of the prebuilt models is selected from the prebuilt model database. For example, the model database contains a prebuilt model A that supports cell line a, a prebuilt model B that supports cell line b, and a prebuilt model C that supports cell lines a and c. If the cell population to be counted belongs to cell line a, prebuilt model A can be selected; if the cell population belongs to cell line c, prebuilt model C can be selected; if the cell population belongs to cell line b', prebuilt model B corresponding to cell line b that is closest to cell line b' is selected.

[0136] S3, determining whether there are defective features in any area of ​​the result image, wherein the defective features include one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality; if so, executing steps S4-S6; otherwise, executing step S6;

[0137] If the current result image does not have defect features, then directly execute S6, that is, based on the current result image without defect features, count the cell information, wherein the cell information includes one or more of the number of cells, the size of cells, and the shape of cells.

[0138] If the current result image has defect features, execute: S4, mark at least one area with defect features, as shown by the two arrows in Figure 9(c) (the arrows may not be displayed on the actual interface); S5, retrain the pre-built model according to the marking operation, and input the target image information into the retrained model to obtain a new result image; in this embodiment, the target image in step S1 and at least the local area image of the marking operation (it can also be the result image after the marking operation) are input into the pre-built model to retrain and obtain a new model; however, the present invention is not limited to the input of the target image in step S1. In another embodiment, the cell group can be re-imaged using a microscope to obtain the current new target image.

[0139] When the target image is a fluorescent cell image, as shown in FIG10(a), it is input into the pre-built model to obtain the result image shown in FIG10(b), which also displays the number of cells. FIG10(c) shows the labeling of the result image, wherein the labeled points identify cells and the labeled lines identify the background (non-cells). The labeling operation and the target image are then input into the current model for retraining to obtain the retrained model. The target image is then input into the retrained model to obtain the new result image shown in FIG10(d). It can be seen that the accuracy of the cell number obtained by the new model recognition is improved.

[0140] S6. Based on the current result image, count the cell information, where the cell information includes one or more of the number of cells, the size of cells, and the shape of cells.

[0141] The execution steps S4-S6 here include multiple situations:

[0142] Case 1: Regardless of whether the new result image obtained in step S5 still has defect features, step S6 is directly executed;

[0143] Case 2: If Figure 2 As shown, after step S5, steps S3 to S5 are repeatedly performed until there is no area with defective features in the result image, and then step S6 is performed;

[0144] Case 3: After step S5, steps S3 to S5 are repeated until a preset number of times, such as 5 times, and then step S6 is executed regardless of whether the new result image obtained in step S5 still has defect features.

[0145] Specifically, in the process of retraining the pre-built model according to the labeling operation in step S5, only the area where the labeling operation is performed is focused on and learned, including: extracting features according to the labeling operation to retrain the model.

[0146] “Focus on and learn only the area where the annotation operation is performed” is also a feature in steps M3 and M305, such as Figure 6 As shown, the focusing and learning of only the area where the annotation operation is performed includes:

[0147] Delimiting the region for the labeling operation so that the defined region includes the cells involved in the labeling operation;

[0148] Generate a sub-image of the defined area, and input the sub-image and corresponding annotation operation information into the model to be retrained as a deep learning sample;

[0149] The corresponding annotation operation information is used as a learning target, and the features of the sub-image are deeply learned to update the neural network parameters in the model to obtain a retrained model.

[0150] Specifically, the annotation operation is an operation performed on defect features in the image, and the annotation operation specifically includes background annotation and / or cell annotation in the form of manual annotation. For example, if two cells in a certain area of ​​the image are manually judged to be adhered, background annotation can be used, that is, a background marking line is drawn between the two cells; if a cell somewhere in the image is manually judged to be ignored (misidentified as the background), cell annotation can be used, that is, a cell marking point is drawn on the cell. In this embodiment, background annotation and cell annotation use different colors to distinguish the annotation type during manual annotation, so as to avoid the misuse of background annotation when marking cells without noticing.

[0151] In one embodiment of the present invention, if there are multiple defect features of different types in the image, it is necessary to mark all the defect features of different types; if there are multiple defect features of the same type in the image, only one of the defect features can be marked, that is, there is no need to force the user to mark too finely, only those places that are obviously unsatisfactory need to be roughly marked, and feedback can be obtained after each marking, using this progressive interactive method until the desired effect is achieved. Then, after the model has performed deep learning on the labeled defect features, the neural network parameters in the model are updated to obtain a retrained model, and the image is input into the retrained model again. If the resulting image shows that not only the previously labeled defect features disappear, but also the defect features of the same type in other places disappear, then the effectiveness of this deep learning is verified to be 100%; if the defect features of the same type in other places have not completely disappeared, the effectiveness of this deep learning is obtained according to a preset standard. For example, if less than one-tenth of the defect features of the same type have not disappeared, the effectiveness of the deep learning is 90%; if more than one-fifth of the defect features of the same type have not disappeared, the effectiveness of the deep learning is unqualified. For unqualified cases, the image can be returned to its initial state to re-label one or more of the defect features.

[0152] Before or after step S6, the method further includes:

[0153] S7. Save the current retrained model and add it to the model database.

[0154] Taking the case where the cell population to be counted belongs to cell line b', a similar pre-built model B is selected in step S2. When step S6 is executed, if the pre-built model is retrained according to the annotation operation, the retrained model can be used as the pre-built model B' that supports cell line b'; if information statistics are to be performed on cell line b' next time, the pre-built model B' can be directly selected.

[0155] Alternatively, taking the case where the cell population to be counted belongs to cell line a, the pre-built model A is selected in step S2, and if the pre-built model is subsequently retrained according to the annotation operation, the retrained model can be used as the optimization model of the pre-built model A that supports cell line a; if information statistics are to be performed on cell line a next time, the optimization model of the pre-built model A can be directly selected. With the iterative update of the model, the learning ability of the optimization model is gradually enhanced, so that the number of defective features in the result image can be gradually reduced during the next execution of steps S1 to S6, further improving the simplicity of operation.

[0156] In one embodiment of the present invention, a corresponding client is provided, and the user performs the following operations on the client to perform cell segmentation:

[0157] The user loads the target image information of the cell group, selects and loads a pre-built model, runs it to get a result image, and if the user is satisfied, there is no need to perform subsequent steps. The current result image is used as the cell segmentation result for counting or other information statistics;

[0158] If the user is not satisfied, the dissatisfied area will be marked, for example, the background between the adhered cells will be marked, and the cells will be marked at the disappeared cells. After the marking is completed, the retraining button of the client will be clicked, and then the client interface will display a new result image. Repeat the above steps until the result image meets the user's expectations, otherwise continue to mark and retrain. Finally, the satisfactory result image is used as the cell segmentation result for counting or other information statistics.

[0159] Cell morphology varies, and different cells have different morphologies under a microscope. On the other hand, researchers may administer drugs to cells or change the culture medium, which may affect the morphology of cells. Furthermore, different light intensities or exposure times have a great impact on imaging, which in turn has a great impact on the algorithm results. In an embodiment of the present invention, users can train and save their own models to meet their own application scenarios. Usually, a user is only interested in a few specific cell lines. They can save their own models, allowing users to train their own models in a very short time, so almost all user needs can be met. In addition, the more times the model is used to identify cell images, the more accurate the model output results will be, and a very high accuracy rate will be achieved in the end. In addition, there is no need to force users to mark too finely, only to roughly mark those areas that are obviously unsatisfactory. Feedback can be obtained after each marking, and this progressive interactive method is used until the desired effect is achieved. In the test of 578 HeLa phase contrast images, the verification results obtained were an average accuracy of 94.42%, R 2 (the proportion of change in the dependent variable predicted from the independent variable) reaches 0.9931.

[0160] In one embodiment of the present invention, a cell information statistics device is provided for performing information statistics on cells in a cell group, such as Figure 7 As shown, the cell information statistics device includes the following modules:

[0161] A target acquisition module to be counted, which is configured to acquire target image information of a cell group to be counted;

[0162] A pre-built model calling module is configured to select a pre-built model and input the target image information into the pre-built model to obtain a result image;

[0163] A defect feature determination module, configured to determine an area in the result image where defect features exist, wherein the defect features include one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality;

[0164] a labeling and retraining module, configured to receive the determination result of the defect feature determination module, label at least one area where the defect feature exists, retrain the pre-built model according to the labeling operation, and input the target image information into the retrained model to obtain a new result image;

[0165] The statistical module is configured to perform statistics on cell information based on the result image, wherein the cell information includes one or more of cell number, cell size, and cell shape.

[0166] like Figure 7 As shown, the pre-built model calling module includes the following units:

[0167] A labeled sample unit, which is configured to obtain multiple images of the cell line and manually label them to obtain labeled learning samples;

[0168] A basic model unit, configured to design a deep learning model and train the deep learning model using the labeled learning samples to obtain a basic model;

[0169] The pre-built model unit is configured to retrain the basic model using a semi-supervised learning method, wherein the semi-supervised learning method obtains the pre-built model by focusing on and learning only the area where the annotation operation is performed.

[0170] like Figure 8 As shown, the pre-built model unit includes the following sub-units:

[0171] A sample image set acquisition subunit, configured to acquire a sample image set for establishing a pre-built model, wherein the sample image set includes a plurality of sample images;

[0172] A traversal subunit, configured to traverse the sample image set and send one of the sample images to the training subunit each time;

[0173] A training subunit, configured to input the received sample image into the basic model to obtain a training result image;

[0174] a defect determination subunit, configured to receive the training result image obtained by the training subunit, and determine whether there is a defect feature in a region of the training result image, wherein the defect feature includes one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality;

[0175] A marking and retraining subunit is configured to mark at least one area with defect features; then retrain the basic model according to the marking operation, only focusing on and learning the area where the marking operation is performed during the retraining process, and obtaining a new training result image based on the retrained current model, and sending it to the defect identification subunit;

[0176] A model updating subunit, configured to update the basic model to the current model when the defect identification subunit identifies a defect-free feature; and trigger the traversal subunit to send the next sample image to the training subunit;

[0177] The model saving subunit is configured to save the current model in response to the completion of the traversal process of the traversal subunit to obtain the pre-built model.

[0178] In one embodiment of the present invention, an electronic device is provided, including a processor and a memory, wherein the memory is used to store program instructions, and the processor is configured to run the program instructions, and the program instructions are run to perform the steps performed in the above method embodiment.

[0179] In one embodiment of the present invention, a computer-readable storage medium is provided for storing program instructions, wherein the program instructions are configured to be called to execute the steps performed in the above method embodiment.

[0180] In one embodiment of the present invention, a computer program product is provided, including a readable and stored computer program, wherein the computer program includes program instructions. When the program instructions are executed on a computer device, the computer device executes the steps executed in the above method embodiment.

[0181] It should be noted that the above-mentioned cell information statistics device, electronic device, computer-readable storage medium, computer program product embodiments and cell information statistics method embodiments belong to the same inventive concept, and the entire content of the cell information statistics method embodiments is incorporated into the cell information statistics device, electronic device, computer-readable storage medium, and computer program product embodiments by reference.

[0182] Some embodiments relate to a microscope including a method and apparatus for statistically analyzing cell information. Optionally, the microscope may be a microscope that performs a combination of Figures 1 to 6 A system or a method of processing one or more of the methods of the invention may be a part of or connected to the system. Fig.11A schematic diagram of a system 100 configured to perform the methods described herein is shown. System 100 includes a microscope 110 and a computer system 120. Microscope 110 is configured to capture images and is connected to computer system 120. Computer system 120 is configured to perform at least a portion of the methods described herein. Computer system 120 can be configured to execute a machine learning algorithm. Computer system 120 and microscope 110 can be separate entities, but can also be integrated into a common housing. Computer system 120 can be part of a central processing system of microscope 110 and / or, computer system 120 can be part of a subcomponent of microscope 110, such as a sensor, actuator, camera or lighting unit of microscope 110.

[0183] The computer system 120 may be a local computer device (e.g., a personal computer, notebook, tablet computer, or mobile phone) having one or more processors and one or more storage devices, or may be a distributed computer system (e.g., having one or more processors and one or more storage devices distributed in various locations, such as a local client and / or one or more remote server sites and / or data centers). The computer system 120 may include any circuit or combination of circuits. In one embodiment, the computer system 120 may include one or more processors that may be of any type. As used in the present invention, a processor may refer to any type of computing circuit, such as, but not limited to, a microprocessor, a microcontroller, a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a graphics processor, a digital signal processor (DSP), a multi-core processor, a field programmable gate array (FPGA) such as a microscope or a microscope component (e.g., a camera), or any other type of processor or processing circuit. Other types of circuits that may be included in the computer system 120 may be custom circuits, application specific integrated circuits (ASICs), etc., such as one or more circuits (e.g., communication circuits) used in wireless devices such as mobile phones, tablet computers, laptops, two-way radios, and similar electronic systems. The computer system 120 may include one or more storage devices, which may include one or more storage elements suitable for a particular application, such as main memory in the form of random access memory (RAM), one or more hard disk drives, and / or one or more drives for handling removable media such as compact disks (CDs), flash memory cards, digital video disks (DVDs), etc. The computer system 120 may also include a display device, one or more speakers, and a keyboard and / or controller, which may include a mouse, trackball, touch screen, voice recognition device, or any other device that allows a system user to input information to or receive information from the computer system 120.

[0184] Some or all of the method steps may be performed by (or using) a hardware device (e.g., a processor, a microprocessor, a programmable computer or an electronic circuit). In some embodiments, such a device may perform one or more of the most important method steps.

[0185] Depending on certain implementation requirements, embodiments of the present invention may be implemented in hardware or software. This implementation may be performed using a non-transitory storage medium (such as a digital storage medium, e.g., a floppy disk, DVD, Blu-ray, CD, ROM, PROM and EPROM, EEPROM or FLASH) on which electronically readable control signals are stored, which cooperate (or can cooperate) with a programmable computer system to perform the corresponding method. Thus, the digital storage medium may be computer readable.

[0186] Some embodiments of the invention comprise a data carrier having electronically readable control signals, which are capable of cooperating with a programmable computer system, such that one of the methods described herein is performed.

[0187] Generally, embodiments of the present invention can be implemented as a computer program product with a program code, when the computer program product runs on a computer, the program code is executable to perform one of the methods. The program code may, for example, be stored on a machine readable carrier.

[0188] Other embodiments comprise the computer program for performing one of the methods described herein, stored on a machine readable carrier.

[0189] In other words, an exemplary embodiment of the present invention is, therefore, a computer program having a program code for performing one of the methods described herein, when the computer program runs on a computer.

[0190] Therefore, a further embodiment of the invention is a storage medium (or a data carrier or a computer-readable medium) comprising a computer program stored thereon for performing one of the methods described in the invention when the computer program is executed by a processor. The data carrier, the digital storage medium or the recorded medium is generally tangible and / or non-transitory. A further embodiment of the invention is an apparatus as described in the invention, comprising a processor and a storage medium.

[0191] Therefore, a further embodiment of the present invention is a data stream or a signal sequence representing the computer program for performing one of the methods described in the present invention. The data stream or the signal sequence can be configured, for example, to be transmitted via a data communication connection, for example via the Internet.

[0192] A further embodiment comprises a processing means, for example a computer or a programmable logic device, configured to or adapted to perform one of the methods described herein.

[0193] A further embodiment comprises a computer on which is installed the computer program for performing one of the methods described herein.

[0194] Another embodiment of the present invention comprises an apparatus or system configured to transmit (for example, electronically or optically) a computer program for performing one of the methods described in the present invention to a receiver. The receiver may be, for example, a computer, a mobile device, a storage device, etc. The apparatus or system may, for example, comprise a file server for transmitting the computer program to the receiver.

[0195] In some embodiments, a programmable logic device (e.g., a field programmable gate array) can be used to perform some or all of the functions of the method described in the present invention. In some embodiments, the field programmable gate array can cooperate with a microprocessor to perform one of the methods described in the present invention. Generally, the method is preferably performed by any hardware device.

[0196] Embodiments may be based on the use of machine learning models or machine learning algorithms. Machine learning may refer to algorithms and statistical models that a computer system can use to perform specific tasks without using explicit instructions, but relying on models and reasoning. For example, in machine learning, instead of rule-based data transformations, data transformations derived from the analysis of historical and / or training data may be used. For example, a machine learning model may be used or a machine learning algorithm may be used to analyze the content of an image. In order for a machine learning model to analyze the content of an image, a machine learning model may be trained using a training image as input and training content information as output. By training a machine learning model with a large number of training images and / or training sequences (e.g., words or sentences) and associated training content information (e.g., labels or annotations), the machine learning model "learns" to recognize the content of the image, so that the machine learning model may be used to recognize image content not included in the training data. The same principle can also be used for other types of sensor data: by training a machine learning model using training sensor data and expected outputs, the machine learning model can "learn" the conversion between sensor data and output, which can be used to provide output based on non-training sensor data provided to the machine learning model. The provided data (e.g., sensor data, metadata, and / or image data) may be preprocessed to obtain a feature vector, which is used as an input to the machine learning model.

[0197] The machine learning model can be trained using training input data. The example specified above uses a training method called "supervised learning". In supervised learning, a machine learning model is trained using multiple training samples, each of which may include multiple input data values, and multiple expected output values, i.e., each training sample is associated with an expected output value. By specifying training samples and expected output values, the machine learning model "learns" the output value to be provided based on input samples similar to the samples provided during the training process. In addition to supervised learning, semi-supervised learning can also be used. In semi-supervised learning, some training samples lack corresponding expected output values. Guided learning can be based on a guided learning algorithm (e.g., a classification algorithm, a regression algorithm, or a similarity learning algorithm). When the output is restricted to a set of restricted values ​​(classification variables), i.e., the input is classified as one of a set of restricted values, a classification algorithm can be used. When the output may have any numerical value (within a range), a regression algorithm can be used. A similarity learning algorithm may be similar to both classification and regression algorithms, but is based on learning from examples using a similarity function that measures the degree of similarity or correlation between two objects. In addition to supervised learning or semi-supervised learning, unsupervised learning can also be used to train a machine learning model. In unsupervised learning, input data may be (only) provided, and an unsupervised learning algorithm may be used to find structure in the input data (e.g., by grouping or clustering the input data to find commonalities in the data). Clustering is the process of assigning input data containing multiple input values ​​to subsets (clusters) such that input values ​​in the same cluster are similar according to one or more (predefined) similarity criteria, but dissimilar to input values ​​contained in other clusters.

[0198] Reinforcement learning is the third group of machine learning algorithms. In other words, reinforcement learning can be used to train machine learning models. In reinforcement learning, one or more software actors (called "software agents") are trained to take actions in an environment. Based on the actions taken, rewards are calculated. Reinforcement learning is based on training one or more software agents to select actions so that the cumulative reward increases, so that the software agents get better at performing the task (as evidenced by increasing rewards).

[0199] In addition, certain techniques can be applied to some machine learning algorithms. For example, feature learning can be used. In other words, feature learning can be used to at least partially train a machine learning model, and / or a machine learning algorithm can include a feature learning component. Feature learning algorithms (which can be called representation learning algorithms) can retain the information in their input, but can also transform it in a way that makes it useful, usually as a preprocessing step before performing classification or prediction. Feature learning can be based on principal component analysis or cluster analysis, for example.

[0200] In some examples, anomaly detection (i.e., outlier detection) can be used, the purpose of which is to provide identification of input values ​​that are suspicious by being significantly different from the majority of inputs or training data. In other words, anomaly detection can be used to at least partially train a machine learning model, and / or a machine learning algorithm can include an anomaly detection component.

[0201] In some examples, a machine learning algorithm may use a decision tree as a prediction model. In other words, a machine learning model may be based on a decision tree. In a decision tree, an observation about an item (e.g., a set of input values) may be represented by a branch of the decision tree, and an output value corresponding to the item may be represented by a leaf of the decision tree. A decision tree may support both discrete and continuous values ​​as output values. If discrete values ​​are used, the decision tree may be represented as a classification tree, and if continuous values ​​are used, the decision tree may be represented as a regression tree.

[0202] Association rules are another technique that can be used in machine learning algorithms. In other words, a machine learning model can be based on one or more association rules. Association rules are created by identifying relationships between variables in a large amount of data. A machine learning algorithm can identify and / or utilize one or more relationship rules that represent knowledge derived from the data. The rules can be used to store, manipulate, or apply this knowledge, for example.

[0203] Machine learning algorithms are often based on machine learning models. In other words, the term "machine learning algorithm" may refer to a set of instructions that can be used to create, train, or use a machine learning model. The term "machine learning model" may refer to a data structure and / or a rule set that represents learned knowledge (e.g., based on training performed by a machine learning algorithm). In an embodiment, the use of a machine learning algorithm may mean the use of a potential machine learning model (or multiple potential machine learning models). The use of a machine learning model may mean that the machine learning model and / or the data structure / rule set that is the machine learning model is trained by a machine learning algorithm.

[0204] For example, a machine learning model can be an artificial neural network (ANN). ANN is a system inspired by biological neural networks, such as those found in the retina or brain. ANN includes multiple interconnected nodes and multiple connections between nodes, so-called edges. There are usually three types of nodes: input nodes that receive input values, hidden nodes that are (only) connected to other nodes, and output nodes that provide output values. Each node can represent an artificial neuron. Each edge can transfer information from one node to another. The output of a node can be defined as a (nonlinear) function of its input (e.g., the sum of its inputs). The input of a node can be used in a function based on the "weight" of the edge or node that provides the input. The weights of nodes and / or edges can be adjusted during the learning process. In other words, the training of an artificial neural network can include adjusting the weights of the nodes and / or edges of the artificial neural network, i.e., achieving the desired output for a given input.

[0205] Alternatively, the machine learning model can be a support vector machine, a random forest model, or a gradient boosting model. A support vector machine (i.e., a support vector network) is a guided learning model with an associated learning algorithm, which can be used to analyze data (e.g., in classification or regression analysis). A support vector machine can be trained by providing multiple training input values ​​belonging to one of two categories to the input. A support vector machine can be trained to assign new input values ​​to one of two categories. Alternatively, the machine learning model can be a Bayesian network, which is a probabilistic directed acyclic graphical model. A Bayesian network can use a directed acyclic graph to represent a set of random variables and their conditional dependencies. Alternatively, the machine learning model can be based on a genetic algorithm, which is a search algorithm and exploration technique that mimics the natural selection process.

[0206] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0207] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items and may be abbreviated as " / ".

[0208] The above description is only a specific implementation of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of this application. These improvements and modifications should also be regarded as the scope of protection of this application.

Claims

1. A method for counting cell information, for counting cell population information, characterized in that: The following steps are involved: S1, obtaining target image information of a cell group to be counted; S2, select a pre-built model, and input the target image information into the pre-built model to obtain a result image; wherein the pre-built model is trained by the following steps: M1, obtain multiple images of the cell line, and manually annotate them to obtain annotated learning samples; M2, design a deep learning model, and use the annotated learning samples to train the deep learning model to obtain a basic model; M3, use a semi-supervised learning method to retrain the basic model, and the semi-supervised learning method adopts a method of focusing on and learning only the area where the annotation operation is performed to obtain the pre-built model; S3, determining whether there are defective features in any area of ​​the result image, wherein the defective features include one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality; if so, executing steps S4-S6; otherwise, executing step S6; S4, marking at least one region with defect features, wherein the marking operation is manually marking the defect features in the image; S5, retraining the pre-built model according to the labeling operation, paying attention to and learning the area where the labeling operation is performed during the retraining process, including: extracting features according to the labeling operation to retrain the model; and inputting the target image information into the retrained model to obtain a new result image; S6. Based on the current result image, count the cell information, where the cell information includes one or more of the number of cells, the size of cells, and the shape of cells.

2. The cell information statistical method according to claim 1, characterized in that: After step S5, steps S3 to S5 are repeatedly performed until a preset number of repetitions is reached, or until there is no area with defective features in the result image, and then step S6 is performed.

3. The cell information statistical method according to claim 1, characterized in that: Step M3 further comprises: M301. Obtain a sample image set for establishing a pre-built model, wherein the sample image set includes a plurality of sample images; M302, input one of the sample images into the basic model to obtain a training result image; M303, determining an area in the training result image where defect features exist, where the defect features include one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality; M304. Mark at least one area with defective features; M305, retraining the current model according to the labeling operation, focusing on and learning only the area where the labeling operation is performed during the retraining process, and obtaining a new training result image based on the retrained current model; Repeat steps M303 to M305 until a preset number of repetitions is reached, or until there is no area with defective features in the training result image, and then execute step M306; M306, updating the basic model to the current model; M307, traverse the sample image set, input the next sample image into the current model to obtain the result image in the current training, and execute steps M303 to M306 until the sample image has been traversed for a preset number of rounds, and then execute M308; M308. Save the current model to obtain the pre-built model.

4. The cell information statistical method according to claim 1, characterized in that: In the process of retraining the pre-built model according to the labeling operation in step S5, only the area where the labeling operation is performed is focused on and learned, including: extracting features according to the labeling operation to retrain the model.

5. The cell information statistical method according to any one of claims 1 to 4, characterized in that: The focusing and learning of only the area where the annotation operation is performed includes: Delimiting the region for the labeling operation so that the defined region includes the cells involved in the labeling operation; Generate a sub-image of the defined area, and input the sub-image and corresponding annotation operation information into the model to be retrained as a deep learning sample; The corresponding annotation operation information is used as a learning target, and the features of the sub-image are deeply learned to update the neural network parameters in the model to obtain a retrained model.

6. The cell information statistical method according to any one of claims 1 to 4, characterized in that: After step M3, the method further includes: M4. Obtain multiple images of other cell lines and manually annotate them to obtain annotated samples of other cell lines; M5. Input the labeled samples of other cell lines into the pre-built model, and retrain it using a semi-supervised learning method to obtain a pre-built model that supports the other cell lines.

7. The cell information statistical method according to any one of claims 1 to 4, characterized in that: The labeling operation includes background labeling and / or cell labeling in the form of manual labeling.

8. The cell information statistical method according to any one of claims 1 to 4, characterized in that: In step S2, one of the pre-built models is selected from the pre-built model database, and before or after step S6, the following steps are further included: S7. Save the current retrained model and add it to the model database.

9. The cell information statistical method according to any one of claims 1 to 4, characterized in that: The target image information of the cell group to be counted in step S1 is one or more of a non-stained and non-fluorescent cell image, a stained cell image, a fluorescent cell image, a phase contrast image, and a bright field image.

10. A cell information statistics device for performing information statistics on cells in a cell group, characterized in that: Includes the following modules: A target acquisition module to be counted, which is configured to acquire target image information of a cell group to be counted; A pre-built model calling module is configured to select a pre-built model and input the target image information into the pre-built model to obtain a result image; wherein the pre-built model calling module includes the following units: a labeling sample unit, which is configured to obtain multiple images of a cell line and manually label them to obtain a labeling learning sample; a basic model unit, which is configured to design a deep learning model and train the deep learning model using the labeling learning sample to obtain a basic model; a pre-built model unit, which is configured to retrain the basic model using a semi-supervised learning method, wherein the semi-supervised learning method obtains the pre-built model by focusing on and learning only the area where the labeling operation is performed; A defect feature determination module, configured to determine an area in the result image where defect features exist, wherein the defect features include one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality; A labeling and retraining module is configured to receive the determination result of the defect feature determination module, perform a labeling operation on at least one area where the defect feature exists, wherein the labeling operation is a manual mark made for the defect feature in the image; and then retrain the pre-built model according to the labeling operation, and pay attention to and learn the area where the labeling operation is performed during the retraining process, including: extracting features according to the labeling operation to retrain the model; and inputting the target image information into the retrained model to obtain a new result image; The statistical module is configured to perform statistics on cell information based on the result image, wherein the cell information includes one or more of cell number, cell size, and cell shape.

11. The cell information counting device according to claim 10, characterized in that: The pre-built model unit includes the following sub-units: A sample image set acquisition subunit, configured to acquire a sample image set for establishing a pre-built model, wherein the sample image set includes a plurality of sample images; A traversal subunit, configured to traverse the sample image set and send one of the sample images to the training subunit each time; A training subunit, configured to input the received sample image into the basic model to obtain a training result image; a defect determination subunit, configured to receive the training result image obtained by the training subunit, and determine whether there is a defect feature in a region of the training result image, wherein the defect feature includes one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality; A marking and retraining subunit, which is configured to perform a marking operation on at least one area having defective features; Then, the basic model is retrained according to the labeling operation, and only the area where the labeling operation is performed is paid attention to and learned during the retraining process, and a new training result image is obtained based on the retrained current model, and the new training result image is sent to the defect discrimination subunit; A model updating subunit, configured to update the basic model to the current model when the defect identification subunit identifies a defect-free feature; and trigger the traversal subunit to send the next sample image to the training subunit; The model saving subunit is configured to save the current model in response to the completion of the traversal process of the traversal subunit to obtain the pre-built model.

12. An electronic device comprising a processor and a memory, wherein: The memory is used to store program instructions, and the processor is configured to execute the program instructions, wherein the program instructions are executed to perform the steps of the method according to any one of claims 1 to 9.

13. A computer-readable storage medium for storing program instructions, characterized in that: The program instructions are configured to be called to execute the steps of the method according to any one of claims 1 to 9.

14. A computer program product, comprising a readable and stored computer program, the computer program comprising program instructions, characterized in that: When the program instructions are executed on a computer device, the computer device executes the steps of the method according to any one of claims 1 to 9.

15. A model training method, characterized in that: The following steps are involved: M1. Obtain multiple images of the cell line and manually annotate them to obtain annotated learning samples; M2. Design a deep learning model, and use the labeled learning samples to train the deep learning model to obtain a basic model; M3. Retrain the basic model using a semi-supervised learning method, wherein the semi-supervised learning method obtains a pre-built model by focusing on and learning only the areas where the annotation operation is performed; Wherein, step M3 further comprises: M301. Obtain a sample image set for establishing a pre-built model, wherein the sample image set includes a plurality of sample images; M302, input one of the sample images into the basic model to obtain a training result image; M303, determining an area in the training result image where defect features exist, where the defect features include one or more of cell adhesion, at least partial disappearance of cells, and cell abnormality; M304, marking at least one region where defect features exist, wherein the marking operation is manually marking the defect features in the image; M305, retraining the current model according to the labeling operation, focusing on and learning only the area where the labeling operation is performed during the retraining process, and obtaining a new training result image based on the retrained current model; Repeat steps M303 to M305 until a preset number of repetitions is reached, or until there is no area with defective features in the training result image, and then execute step M306; M306, updating the basic model to the current model; M307, traverse the sample image set, input the next sample image into the current model to obtain the result image in the current training, and execute steps M303 to M306 until the sample image has been traversed for a preset number of rounds, and then execute M308; M308. Save the current model to obtain the pre-built model.

Images