Method and device for identifying malignant cells

Abstract

The invention provides a malignant cell identification method and device, and the method comprises the steps: obtaining a to-be-detected specimen image which is acquired based on a rapid on-site evaluation technology; the to-be-detected sample image is input into a malignant cell recognition model, a malignant cell recognition result output by the malignant cell recognition model is obtained, and the malignant cell recognition result comprises the number of malignant cells included in the to-be-detected sample image and the position of each malignant cell in the to-be-detected sample image; the malignant cell identification model is obtained by training an initial malignant cell identification model based on a data enhanced sample set, and the real label of a sample image in the data enhanced sample set is the real number of malignant cells and the real position of each malignant cell included in a target sample image; the initial malignant cell recognition model is obtained by taking a YOLOv5 model as a basic model and lightening model parameters of the YOLOv5 model.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a malignant cell identification method and device. Background Art

[0002] In recent years, the research on deep learning has continued to deepen, and the application of deep learning in the identification of malignant cells in cell smears has also made significant progress. Compared with the traditional manual identification methods, which have the disadvantages of strong subjectivity, low efficiency, and easy fatigue, the use of deep learning technology to automatically and intelligently identify malignant cells has become an important research direction.

[0003] However, current technologies still face some shortcomings in practical applications. First, although the model can achieve high recognition accuracy on specific data sets, its generalization ability remains a challenge. There may be differences in cell smears prepared by different processes, resulting in unstable performance of the model in new environments. Secondly, deep learning models have high requirements on the quality and quantity of data, and current data sets are difficult to obtain on a large scale, which limits the training effect and practical application scope of the model. Finally, most current technologies can only give an identification result of whether a single cell image contains malignant cells, but cannot provide further detailed information. This limitation also brings inconvenience in practical applications. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention provides a method and device for identifying malignant cells.

[0005] The present invention provides a method for identifying malignant cells, comprising: Acquiring an image of a sample to be detected, wherein the image of the sample to be detected is collected based on a rapid on-site evaluation technology; Inputting the sample image to be detected into a malignant cell recognition model to obtain a malignant cell recognition result output by the malignant cell recognition model, wherein the malignant cell recognition result includes the number of malignant cells included in the sample image to be detected and the position of each malignant cell in the sample image to be detected; The malignant cell recognition model is obtained by training the initial malignant cell recognition model based on a data-enhanced sample set, and the true labels of the sample images in the data-enhanced sample set are the true number of malignant cells and the true position of each malignant cell included in the target sample image; the initial malignant cell recognition model is obtained by lightweighting the model parameters of the YOLOv5 model based on the YOLOv5 model.

[0006] According to a malignant cell identification method provided by the present invention, the data enhanced sample set is obtained by the following method: Randomly select a number of original sample image groups from the original sample set, each of which includes a number of original sample images; For each of the original sample image groups, combining a plurality of original sample images in the original sample image group into a composite image; Cropping a cropped sample image having the same size as the original sample image from the composite image; Based on the original sample image and the cropped sample image, a data enhanced sample set is obtained.

[0007] According to a malignant cell identification method provided by the present invention, the data enhanced sample set is obtained by the following method: The original sample image in the original sample set is divided into several sub-images; Ensure that the aspect ratio of each sub-image remains unchanged, increase the size of each sub-image, and obtain a slice-enhanced image; A data enhanced sample set is obtained based on the original sample image and the slice enhanced image.

[0008] According to a malignant cell identification method provided by the present invention, the data enhanced sample set is obtained by the following method: Performing geometric and color transformation processing on the original sample images in the original sample set to obtain geometric and color transformed images, wherein the geometric and color transformation processing includes at least one of random cropping, flipping, rotation and brightness transformation; A data-enhanced sample set is obtained based on the original sample image and the image after the geometry and color transformation.

[0009] According to a malignant cell identification method provided by the present invention, the initial malignant cell identification model is obtained by the following method: Taking the YOLOv5-s model in the YOLOv5 model as the basic model, the backbone network in the YOLOv5-s model is replaced with a lightweight neural network model to obtain an initial malignant cell recognition model, wherein the lightweight neural network model includes a MobileNetV3 neural network model or a ShuffleNetV2 neural network model.

[0010] According to a malignant cell identification method provided by the present invention, the initial malignant cell identification model is obtained by the following method: The YOLOv5-n model in the YOLOv5 model is used as a basic model, and the BottleNeck module in the backbone network of the YOLOv5-n model is replaced with a Vision Transformer module to obtain an initial malignant cell recognition model.

[0011] According to a malignant cell recognition method provided by the present invention, the malignant cell recognition model is obtained by training an initial malignant cell recognition model based on a data-enhanced sample set using an average precision mean as an evaluation index.

[0012] The present invention also provides a malignant cell identification device, comprising: A data acquisition module, used to acquire an image of a sample to be detected, wherein the image of the sample to be detected is acquired based on a rapid on-site evaluation technology; an identification module, used for inputting the sample image to be detected into a malignant cell identification model to obtain a malignant cell identification result output by the malignant cell identification model, wherein the malignant cell identification result includes the number of malignant cells included in the sample image to be detected and the position of each malignant cell in the sample image to be detected; The malignant cell recognition model is obtained by training the initial malignant cell recognition model based on a data-enhanced sample set, and the true labels of the sample images in the data-enhanced sample set are the true number of malignant cells and the true position of each malignant cell included in the target sample image; the initial malignant cell recognition model is obtained by lightweighting the model parameters of the YOLOv5 model based on the YOLOv5 model.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the malignant cell identification methods described above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the malignant cell identification method described in any one of the above is implemented.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned malignant cell identification methods.

[0016] The malignant cell identification method and device provided by the present invention can not only identify whether a malignant cell is contained, but also identify detailed information of the malignant cell by adopting a malignant cell identification model improved based on the YOLOv5 model. By introducing data enhancement technology to construct a sample set, the generalization ability of the model is improved. In addition, the present invention also reduces the complexity and calculation amount of the model by lightweight processing of the model parameters of the YOLOv5 model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A schematic diagram of the process of the malignant cell identification method provided by the present invention; Figure 2 One of the schematic diagrams of the data enhancement scenario provided by the present invention; Figure 3 The second schematic diagram of the data enhancement scenario provided by the present invention; Figure 4 A schematic diagram of the structure of the malignant cell identification device provided by the present invention; Figure 5 This is a schematic structural diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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 are within the scope of protection of the present invention.

[0020] Figure 1 A schematic diagram of the process of the malignant cell identification method provided by the present invention is shown in FIG. Figure 1 As shown, the present invention provides a method for identifying malignant cells, comprising: Step 110, obtaining an image of a sample to be detected, wherein the image of the sample to be detected is collected based on a rapid on-site evaluation technology; Here, Rapid On-Site Evaluation (ROSE) can prepare the collected cytological specimens immediately at the operation site, including steps such as pressing, brushing / smearing, and spraying. Rapid staining technology (such as Diff-Quik staining) is used to stain the specimens for observation under a microscope.

[0021] Specifically, during the ROSE process, the cell image observed and photographed through a microscope is the image of the specimen to be detected.

[0022] Step 120, inputting the sample image to be detected into a malignant cell recognition model to obtain a malignant cell recognition result output by the malignant cell recognition model, wherein the malignant cell recognition result includes the number of malignant cells included in the sample image to be detected and the position of each malignant cell in the sample image to be detected.

[0023] In this embodiment, the sample image to be detected is input into the trained malignant cell recognition model. The malignant cell recognition model outputs the malignant cell recognition result. The malignant cell recognition result usually includes two key information: one is the number of malignant cells in the image, and the other is the specific location of each malignant cell in the image, which is usually marked in the form of center point coordinates and bounding box width and height.

[0024] The malignant cell recognition model is obtained by training the initial malignant cell recognition model based on a data-enhanced sample set, and the true labels of the sample images in the data-enhanced sample set are the true number of malignant cells and the true position of each malignant cell included in the target sample image; the initial malignant cell recognition model is obtained by lightweighting the model parameters of the YOLOv5 model based on the YOLOv5 model.

[0025] In this embodiment, it is first necessary to construct a data set containing a large number of sample images. These data sets increase the diversity and complexity of samples through data enhancement techniques such as rotation, scaling, random cropping, flipping, adding noise, etc., thereby improving the generalization ability of the model. The true label of each sample image includes the true number of malignant cells in the image and the true location of each malignant cell, such as through manually annotated bounding boxes.

[0026] In one example, the data augmented sample set is obtained in the following manner: Performing geometric and color transformation processing on the original sample images in the original sample set to obtain geometric and color transformed images, wherein the geometric and color transformation processing includes at least one of random cropping, flipping, rotation and brightness transformation; A data-enhanced sample set is obtained based on the original sample image and the image after the geometry and color transformation.

[0027] refer to Figure 2As shown, in the original image a, a cropping frame is generated, whose position, width and height are randomly generated, and the pixels within the cropping frame area are used as the new image b. Each pixel of the original image a can be swapped with the pixels relative to its horizontal axis or vertical axis to obtain a new image c. Each pixel of the original image a can be rotated by an angle along the center point of the image to obtain a new image d. The brightness of the original image a can also be randomly modified. The color of the image can be represented by three parameters: hue, saturation, and brightness. The hue determines the color of the picture, the saturation determines the purity of the color, and the brightness determines the brightness of the image. For example, by modifying the brightness of the original image a, a new image e taken under different light and dark conditions can be simulated.

[0028] It should be noted that the YOLOv5 model is implemented using the PyTorch framework, which is optimized in terms of speed and accuracy, with faster reasoning speed and better detection performance. In addition, the model provides a wealth of configuration parameters and interfaces to facilitate custom adjustments to the network structure and training parameters, and is suitable for a variety of application scenarios. The YOLOv5 model is mainly composed of four parts, namely the input end Input, the backbone network Backbone, the neck Neck, and the detection end Head. These parts are connected in series to work together

[0029] Specifically, YOLOv5 adds a Focus structure to the backbone network Backbone, and crops the input image through slicing operations. In addition, YOLOv5 designs a new CSP structure and also adds a CSP structure to the neck. In the neck part, the YOLOv5 model uses Feature Pyramid Networks (FPN) and Path Aggregation Network (PAN), and adds the CSP structure in the backbone network Backbone, further enhancing the ability of feature fusion from bottom to top and top to bottom, thereby improving the model's ability to detect and locate targets. In the output layer Head, YOLOv5 provides three output layers, corresponding to three different resolutions. The high-resolution layer can detect small targets, and the low-resolution layer can detect large targets.

[0030] In the embodiment of the present invention, the YOLOv5 model is selected as the basic model, and further, the model parameters of the YOLOv5 model are lightweighted, such as reducing the number of model layers, reducing the number of convolution kernels, using more efficient convolution operations, etc., so as to reduce the complexity and amount of calculation while maintaining the performance of the model.

[0031] The malignant cell recognition method proposed in the embodiment of the present invention can not only identify whether a malignant cell is contained, but also identify detailed information of the malignant cell by adopting a malignant cell recognition model improved based on the YOLOv5 model. By introducing data enhancement technology to construct a sample set, the generalization ability of the model is improved. In addition, the present invention also reduces the complexity and calculation amount of the model by lightweight processing of the model parameters of the YOLOv5 model.

[0032] Based on the above embodiment, the data enhanced sample set is obtained in the following way: Randomly select a number of original sample image groups from the original sample set, each of which includes a number of original sample images; For each of the original sample image groups, combining a plurality of original sample images in the original sample image group into a composite image; Cropping a cropped sample image having the same size as the original sample image from the composite image; Based on the original sample image and the cropped sample image, a data enhanced sample set is obtained.

[0033] refer to Figure 3 As shown, the original sample image group includes 4 original sample images, each of which has the same size. Then, a center point is randomly selected on the stitching canvas as the reference point for stitching the 4 original sample images. According to the center point, the 4 original sample images are placed at the upper left, upper right, lower left, and lower right positions of the stitching canvas, respectively, and stitched into a large-size composite image. Then, a cropped sample image with the same size as the original sample image is cropped from the composite image.

[0034] In the embodiment of the present invention, new sample images are generated by cropping and splicing the original sample images, thereby increasing the scale and diversity of the training data set and improving the generalization ability and accuracy of the model.

[0035] Based on the above embodiment, the data enhanced sample set is obtained in the following way: The original sample image in the original sample set is divided into several sub-images; Ensure that the aspect ratio of each sub-image remains unchanged, increase the size of each sub-image, and obtain a slice-enhanced image; A data enhanced sample set is obtained based on the original sample image and the slice enhanced image.

[0036] In this embodiment, a plurality of overlapping sub-images may be firstly cut out from each original sample image. The sizes of these sub-images may be different, but the width and height of the sub-images must be within a preset range to ensure that there is an overlapping area between the sub-images. Then, under the condition of ensuring that the aspect ratio remains unchanged, the size of each sub-image is increased, for example, the width of the sub-image is adjusted from 864 pixels to between 800 and 1333 pixels, to obtain a slice-enhanced image.

[0037] In the embodiment of the present invention, a new sample image is generated by performing a slice enhancement operation on the original sample image, thereby increasing the scale and diversity of the training data set and improving the generalization ability and accuracy of the model.

[0038] Based on the above embodiment, the initial malignant cell recognition model is obtained in the following manner: Taking the YOLOv5-s model in the YOLOv5 model as the basic model, the backbone network in the YOLOv5-s model is replaced with a lightweight neural network model to obtain an initial malignant cell recognition model, wherein the lightweight neural network model includes a MobileNetV3 neural network model or a ShuffleNetV2 neural network model.

[0039] Here, the YOLOv5-s model refers to the s version of the different versions provided by the YOLOv5 model. The YOLOv5-s model is a smaller version of the YOLOv5 model with fewer parameters and a lighter model.

[0040] In this embodiment, the YOLOv5-s model is used as the basic model, and the backbone network in the YOLOv5-s model is replaced with a lightweight neural network model with fewer parameters, such as a MobileNetV3 neural network model or a ShuffleNetV2 neural network model.

[0041] The MobileNetV3 model adopts lightweight depth-wise separable convolution, residual block and SE channel attention mechanism, achieving good performance and speed.

[0042] Depthwise separable convolution decomposes the convolution operation of traditional CNN into two parts: channel-by-channel convolution (DW) and pointwise convolution (PW). The depthwise separable convolution combined with DW and PW has less computational complexity than traditional convolution. The SE module is similar to channel attention, which extracts the weights of different channels through global pooling and fully connected layers.

[0043] In addition, the MobileNetV3 model also proposes the h-swish activation function, which combines the h-sigmoid function and the swish function. The h-swish function retains the advantages of the swish function, which can alleviate the gradient disappearance and improve the training efficiency of the model. At the same time, the h-sigmoid function is used to replace the sigmoid function, making the calculation process simpler without significantly reducing the performance of the swish function, further reducing the calculation cost, thereby improving the training efficiency.

[0044] The ShuffleNetV2 model is also a lightweight network model that aims to significantly reduce the size of the model and speed up the model's operation without sacrificing too much model performance. The ShuffleNetV2 model consists of multiple convolution blocks, each of which includes a downsampling unit and multiple basic units. The basic unit uses grouped convolution and channel rearrangement technology to reduce the amount of calculation and improve feature fusion capabilities.

[0045] Based on the above embodiment, the initial malignant cell recognition model is obtained in the following manner: The YOLOv5-n model in the YOLOv5 model is used as a basic model, and the BottleNeck module in the backbone network of the YOLOv5-n model is replaced with a Vision Transformer module to obtain an initial malignant cell recognition model.

[0046] Here, the YOLOv5-n model is a version of the YOLOv5 series that is optimized for edge computing. It provides accuracy suitable for edge devices while maintaining a relatively fast speed.

[0047] In this embodiment, the feature extraction network of the YOLOv5-n model includes several C3 modules, in which a structure similar to a residual network is adopted, and several BottleNeck modules are used to first reduce the dimension of the feature, then increase the dimension, and merge the result with the result of only one convolution operation. In this embodiment, the BottleNeck module for dimension reduction and dimension increase is replaced by the Vision Transformer module.

[0048] Here, the Vision Transformer module includes an image segmentation flattening layer, a Patch Embedding, a Position Embedding, a Transformer Encoder, and an MLP Head classification layer. In this embodiment, after the BottleNeck module in the backbone network of the YOLOv5-n model is replaced with the Vision Transformer module, the MLP Head classification layer in the Vision Transformer module is removed.

[0049] On the basis of the above-mentioned embodiment, the malignant cell recognition model is obtained by training the initial malignant cell recognition model based on the data-enhanced sample set and taking the average precision mean as the evaluation index.

[0050] In this embodiment, the mean average precision is selected as the evaluation indicator during model training.

[0051] Specifically, for each category (i.e., malignant cells or non-malignant cells) output by the initial malignant cell recognition model, first draw its precision-recall curve, then calculate the area under the curve, that is, get the average precision value of each category, and average the average precision values ​​of all categories to get the average precision mean. Then, according to the evaluation results of the average precision mean, tune the model.

[0052] The malignant cell identification device provided by the present invention is described below. The malignant cell identification device described below and the malignant cell identification method described above can be referred to each other.

[0053] Figure 4 The schematic diagram of the structure of the malignant cell identification device provided by the present invention is as follows: Figure 4 As shown, the present invention provides a malignant cell identification device, including a data acquisition module 410 and an identification module 420.

[0054] The data acquisition module 410 is used to acquire the image of the sample to be detected, wherein the image of the sample to be detected is acquired based on the rapid on-site evaluation technology; The recognition module 420 is used to input the sample image to be detected into the malignant cell recognition model to obtain the malignant cell recognition result output by the malignant cell recognition model, wherein the malignant cell recognition result includes the number of malignant cells included in the sample image to be detected and the position of each malignant cell in the sample image to be detected; The malignant cell recognition model is obtained by training the initial malignant cell recognition model based on a data-enhanced sample set, and the true labels of the sample images in the data-enhanced sample set are the true number of malignant cells and the true position of each malignant cell included in the target sample image; the initial malignant cell recognition model is obtained by lightweighting the model parameters of the YOLOv5 model based on the YOLOv5 model.

[0055] The malignant cell identification device provided by the present invention can not only identify whether a malignant cell is contained, but also identify detailed information of the malignant cell by adopting a malignant cell identification model improved based on the YOLOv5 model. By introducing data enhancement technology to construct a sample set, the generalization ability of the model is improved. In addition, the present invention also reduces the complexity and calculation amount of the model by lightweight processing of the model parameters of the YOLOv5 model.

[0056] The device provided by the present invention is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for the specific processes and detailed contents, which will not be repeated here.

[0057] Figure 5 A schematic diagram of the structure of an electronic device provided by the present invention, such as Figure 5 As shown, the electronic device may include: a processor 501, a communications interface 502, a memory 503 and a communication bus 504, wherein the processor 501, the communications interface 502 and the memory 503 communicate with each other via the communication bus 504. The processor 501 may call the logic instructions in the memory 503 to execute the malignant cell identification method, which includes: Acquiring an image of a sample to be detected, wherein the image of the sample to be detected is collected based on a rapid on-site evaluation technology; Inputting the sample image to be detected into a malignant cell recognition model to obtain a malignant cell recognition result output by the malignant cell recognition model, wherein the malignant cell recognition result includes the number of malignant cells included in the sample image to be detected and the position of each malignant cell in the sample image to be detected; The malignant cell recognition model is obtained by training the initial malignant cell recognition model based on a data-enhanced sample set, and the true labels of the sample images in the data-enhanced sample set are the true number of malignant cells and the true position of each malignant cell included in the target sample image; the initial malignant cell recognition model is obtained by lightweighting the model parameters of the YOLOv5 model based on the YOLOv5 model.

[0058] In addition, the logic instructions in the above-mentioned memory 503 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for 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 invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0059] On the other hand, the present invention further provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, when the program instructions are executed by a computer, the computer can execute the malignant cell identification method provided by the above methods, the method comprising: Acquiring an image of a sample to be detected, wherein the image of the sample to be detected is collected based on a rapid on-site evaluation technology; Inputting the sample image to be detected into a malignant cell recognition model to obtain a malignant cell recognition result output by the malignant cell recognition model, wherein the malignant cell recognition result includes the number of malignant cells included in the sample image to be detected and the position of each malignant cell in the sample image to be detected; The malignant cell recognition model is obtained by training the initial malignant cell recognition model based on a data-enhanced sample set, and the true labels of the sample images in the data-enhanced sample set are the true number of malignant cells and the true position of each malignant cell included in the target sample image; the initial malignant cell recognition model is obtained by lightweighting the model parameters of the YOLOv5 model based on the YOLOv5 model.

[0060] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the malignant cell identification method provided in the above embodiments is implemented, and the method includes: Acquiring an image of a sample to be detected, wherein the image of the sample to be detected is collected based on a rapid on-site evaluation technology; Inputting the sample image to be detected into a malignant cell recognition model to obtain a malignant cell recognition result output by the malignant cell recognition model, wherein the malignant cell recognition result includes the number of malignant cells included in the sample image to be detected and the position of each malignant cell in the sample image to be detected; The malignant cell recognition model is obtained by training the initial malignant cell recognition model based on a data-enhanced sample set, and the true labels of the sample images in the data-enhanced sample set are the true number of malignant cells and the true position of each malignant cell included in the target sample image; the initial malignant cell recognition model is obtained by lightweighting the model parameters of the YOLOv5 model based on the YOLOv5 model.

[0061] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0062] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying malignant cells, characterized in that: include: Acquiring an image of a sample to be detected, wherein the image of the sample to be detected is collected based on a rapid on-site evaluation technology; Inputting the sample image to be detected into a malignant cell recognition model to obtain a malignant cell recognition result output by the malignant cell recognition model, wherein the malignant cell recognition result includes the number of malignant cells included in the sample image to be detected and the position of each malignant cell in the sample image to be detected; The malignant cell recognition model is obtained by training the initial malignant cell recognition model based on a data-enhanced sample set, and the true labels of the sample images in the data-enhanced sample set are the true number of malignant cells and the true position of each malignant cell included in the target sample image; the initial malignant cell recognition model is obtained by lightweighting the model parameters of the YOLOv5 model based on the YOLOv5 model.

2. The malignant cell identification method according to claim 1, characterized in that: The data augmentation sample set is obtained in the following way: Randomly select a number of original sample image groups from the original sample set, each of which includes a number of original sample images; For each of the original sample image groups, combining a plurality of original sample images in the original sample image group into a composite image; Cropping a cropped sample image having the same size as the original sample image from the composite image; Based on the original sample image and the cropped sample image, a data enhanced sample set is obtained.

3. The malignant cell identification method according to claim 1, characterized in that: The data augmentation sample set is obtained in the following way: The original sample image in the original sample set is divided into several sub-images; Ensure that the aspect ratio of each sub-image remains unchanged, increase the size of each sub-image, and obtain a slice-enhanced image; A data enhanced sample set is obtained based on the original sample image and the slice enhanced image.

4. The malignant cell identification method according to claim 1, characterized in that: The data augmentation sample set is obtained in the following way: Performing geometric and color transformation processing on the original sample images in the original sample set to obtain geometric and color transformed images, wherein the geometric and color transformation processing includes at least one of random cropping, flipping, rotation and brightness transformation; Based on the original sample image and the image after the geometric and color transformation, a data enhanced sample set is obtained.

5. The malignant cell identification method according to claim 1, characterized in that: The initial malignant cell recognition model is obtained in the following manner: Taking the YOLOv5-s model in the YOLOv5 model as the basic model, the backbone network in the YOLOv5-s model is replaced with a lightweight neural network model to obtain an initial malignant cell recognition model, wherein the lightweight neural network model includes a MobileNetV3 neural network model or a ShuffleNetV2 neural network model.

6. The malignant cell identification method according to claim 1, characterized in that: The initial malignant cell recognition model is obtained in the following manner: The YOLOv5-n model in the YOLOv5 model is used as a basic model, and the BottleNeck module in the backbone network of the YOLOv5-n model is replaced with a Vision Transformer module to obtain an initial malignant cell recognition model.

7. The malignant cell identification method according to claim 1, characterized in that: The malignant cell recognition model is obtained by training the initial malignant cell recognition model based on a data-enhanced sample set and taking the average precision mean as an evaluation index.

8. A malignant cell identification device, characterized in that: include: A data acquisition module, used to acquire an image of a sample to be detected, wherein the image of the sample to be detected is acquired based on a rapid on-site evaluation technology; an identification module, used for inputting the sample image to be detected into a malignant cell identification model to obtain a malignant cell identification result output by the malignant cell identification model, wherein the malignant cell identification result includes the number of malignant cells included in the sample image to be detected and the position of each malignant cell in the sample image to be detected; The malignant cell recognition model is obtained by training the initial malignant cell recognition model based on a data-enhanced sample set, and the true labels of the sample images in the data-enhanced sample set are the true number of malignant cells and the true position of each malignant cell included in the target sample image; the initial malignant cell recognition model is obtained by lightweighting the model parameters of the YOLOv5 model based on the YOLOv5 model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the malignant cell identification method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the malignant cell identification method according to any one of claims 1 to 7 is implemented.

Images