Image processing method, apparatus and device

Abstract

The application discloses an image processing method, device and equipment, which can be applied to various scenes such as artificial intelligence medical technology and machine learning. The method comprises the following steps: acquiring a plurality of field-of-view images of a target digital slice; extracting cell image features and a positive probability corresponding to a positive cell in each field-of-view image based on a cell detection model; determining a target positive cell in the target digital slice based on the positive probability corresponding to the positive cell in the field-of-view image; and when the positive probability of the target positive cell matches a suspicious probability interval, determining a slice category of the target digital slice based on the cell detection model and the cell image features and the positive probability corresponding to the positive cell, thereby assisting a pathologist in cytological diagnosis, reducing the workload of the pathologist and improving the efficiency and accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to an image processing method, device and equipment. BACKGROUND

[0002] In the field of medical health, doctors can observe cell pictures through microscopic equipment for pathological diagnosis. However, due to the lack of cytological pathologists in primary hospitals, and due to the inconsistency of pathologists and high cost of medical education, there have been extensive researches to assist pathologists in diagnosis through artificial intelligence technology.

[0003] However, in the related art, most of the negative cells in the cell section can only be excluded, and the doctor still needs to manually check the cell section to make a judgment, and there is the shortcoming of low accuracy. SUMMARY

[0004] The embodiments of the present application provide an image processing method, device and equipment, which improves the accuracy of pathological cell classification and reduces the workload of pathologists.

[0005] In one aspect, an image processing method is provided, applied to a computer device, wherein the computer device is deployed with a cascaded cell detection model and a whole section classification model, and the method comprises:

[0006] obtaining a plurality of field images of a target digital section;

[0007] extracting, based on the cell detection model, cell image features and positive probabilities corresponding to positive cells in each of the field images;

[0008] determining a target positive cell in the target digital section based on the positive probabilities corresponding to the positive cells in the field images;

[0009] when the positive probability corresponding to the target positive cell matches a suspicious probability interval, determining a section category of the target digital section based on the whole section classification model according to the cell image features and the positive probability corresponding to the positive cell.

[0010] In another aspect, an image processing device is provided, applied to a computer device, wherein the computer device is deployed with a cascaded cell detection model and a whole section classification model, and the device comprises:

[0011] a first obtaining module configured to obtain a plurality of field images of a target digital section;

[0012] a second obtaining module configured to extract, based on the cell detection model, cell image features and positive probabilities corresponding to positive cells in each of the field images;

[0013] The first determining module is configured to determine a target positive cell in the target digital slice based on the positive probability corresponding to the positive cell in the field-of-view image.

[0014] The second determining module is configured to determine the slice category of the target digital slice based on the cell image feature and the positive probability corresponding to the positive cell according to the full-slide classification model when the positive probability corresponding to the target positive cell matches the suspicious probability interval.

[0015] In another aspect, a computer readable storage medium is provided, which stores a computer program adapted to be loaded into a processor to perform the steps in the image processing method according to any one of the above embodiments.

[0016] In another aspect, a computer device is provided, which includes a processor and a memory, and the memory stores a computer program, and the processor is configured to perform the steps in the image processing method according to any one of the above embodiments by invoking the computer program stored in the memory.

[0017] In another aspect, a computer program product is provided, which includes computer instructions for performing the steps in the image processing method according to any one of the above embodiments when executed by a processor.

[0018] Embodiments of the present application provide an image processing method, device and equipment, which obtains a plurality of field-of-view images of a target digital slice, then extracts a cell image feature and a positive probability corresponding to a positive cell in each field-of-view image based on a cell detection model, then determines a target positive cell in the target digital slice based on the positive probability corresponding to the positive cell in the field-of-view image, and when the positive probability corresponding to the target positive cell matches a suspicious probability interval, determines the slice category of the target digital slice based on a full-slide classification model according to the cell image feature and the positive probability corresponding to the positive cell. Embodiments of the present application cascade the cell detection model and the full-slide classification model, detect a positive cell in the target digital slice based on the cell detection model, obtain a feature map and a positive probability corresponding to the positive cell, determine the target positive cell first, and determine the slice category of the target digital slice according to the positive probability corresponding to the target positive cell for the first time, and if the positive probability corresponding to the target positive cell matches the suspicious probability interval, determine the slice category of the target digital slice based on the full-slide classification model, thereby judging the pathological category of the target digital slice, assisting a pathologist in cytological diagnosis, reducing the workload of the pathologist, and improving the efficiency and accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort based on these drawings.

[0020] Figure 1 The first flowchart of the image processing method provided by the embodiments of the present application.

[0021] Figure 2 The first application scenario diagram of the image processing method provided by the embodiments of the present application.

[0022] Figure 3 The second application scenario diagram of the image processing method provided by the embodiments of the present application.

[0023] Figure 4 The third application scenario diagram of the image processing method provided by the embodiments of the present application.

[0024] Figure 5 The second flowchart of the image processing method provided by the embodiments of the present application.

[0025] Figure 6 The structural diagram of the image processing device provided by the embodiments of the present application.

[0026] Figure 7 The structural diagram of the computer device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort belong to the scope of protection of the present application.

[0028] Embodiments of the present application provide an image processing method and device, computer equipment and a storage medium. Specifically, the image processing method of the embodiments of the present application can be executed by computer equipment, which can be a terminal or a server, etc. The terminal can be a smart phone, a tablet computer, a notebook computer, a smart television, a smart speaker, a wearable smart device, a smart vehicle terminal, etc. The terminal can also include a client, which can be an application program client, a browser client or an instant messaging client, etc. The server can be a standalone physical server, a server cluster or a distributed system composed of multiple physical servers, a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and basic cloud computing services such as big data and artificial intelligence platforms, etc.

[0029] The embodiments of the present application can be applied to various scenarios, including but not limited to artificial intelligence medical technology, machine learning, etc. The artificial intelligence medical technology scenario can include medical diagnosis and other application scenarios.

[0030] First, some of the nouns or terms that appear in the description of the embodiments of the present application are explained as follows:

[0031] Artificial intelligence (AI) is the use of digital computers or digital computer-controlled machines to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which aims to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0032] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software technologies. Artificial intelligence basic technologies generally include sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc.

[0033] Computer Vision (CV) Computer vision is a scientific discipline that deals with enabling machines to "see". More formally, it deals with enabling computers to see and understand the world in the same way that humans do. As a scientific discipline, computer vision research is aimed at understanding the real world through the interpretation of images. This field has deep roots in engineering and cognitive science. Computer vision technology usually includes image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, and other technologies. It also includes common face recognition, fingerprint recognition, and other biometric identification technologies.

[0034] Machine Learning (ML) Machine learning is a multidisciplinary field that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, and other disciplines. It is a field that studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent. It is applied in various fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and inductive learning.

[0035] Residual Network (ResNet) is a kind of convolutional neural network. The characteristic of residual network is easy to optimize, and can improve the accuracy by increasing the depth.

[0036] Faster-RCNN (a target detection algorithm) was published in the paper "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks" in 2015. The biggest innovation of this algorithm is the RPN (Region Proposal Network) network, which uses the anchor mechanism to link the region generation and the convolutional network together.

[0037] Whole Slide Image (WSI) is a high-resolution digital pathology image. After a pathology section is digitized, a WSI is formed. The vast amount of information it contains provides a reliable basis for quantitative analysis of digital pathology.

[0038] Positive and negative: In medical examinations, negative generally represents normal and positive represents a problem. Negative and positive are more commonly used in medicine and have become a term to generally refer to the presence or absence or to indicate the results of a certain examination. Generally, positive mostly indicates disease or occurrence of certain pathological changes; negative mostly denies or excludes the possibility of certain pathological changes.

[0039] Cervical cancer is one of the most common malignant tumors threatening women's health, which is concentrated in economically underdeveloped developing countries. The cause of cervical cancer is clear, and it can be effectively reduced through screening, early diagnosis and early treatment. At present, effective screening programs include high-risk HPV detection, cervical cytology screening and combined screening of the two modes. Cervical cytology screening is based on pathological diagnosis of cervical cell smears, which belongs to cell morphology diagnosis.

[0040] In the related art, a neural network model can be used to assist pathologists in pathological diagnosis, however, at present, most of the negative cells in the slice can be excluded to reduce the workload of pathologists, and the doctors still need to check each positive slice to make a diagnosis. In the related whole slice analysis technology, there is still a problem of low accuracy.

[0041] Therefore, an image processing method is provided in the embodiments of the present application, which cascades a cell detection model and a whole slice classification model, detects positive cells in a target digital slice based on the cell detection model, obtains a feature map of the positive cells and a positive probability of the positive cells, determines the target digital slice as a suspicious positive slice according to the positive probability for the first time, determines a slice positive probability of the target digital slice through the whole slice classification model, and determines the slice category of the target digital slice according to the slice positive probability for the second time, thereby judging the pathological category of the target digital slice, assisting pathologists in cytological diagnosis, reducing the workload of doctors, and improving efficiency and accuracy.

[0042] The following will be described in detail. It should be noted that the description order of the following embodiments is not limited as the priority order of the embodiments.

[0043] The embodiments of the present application provide an image processing method, which can be executed by a terminal or a server, or jointly executed by a terminal and a server. The embodiments of the present application take the image processing method executed by the server as an example for description.

[0044] Please refer to Figures 1 to 5 , Figure 1 , Figure 5 are flowcharts of the image processing method provided by the embodiments of the present application, Figures 2 to 4Fig. 1 and Fig. 2 are schematic diagrams of application scenarios of the image processing method provided in the embodiments of the present application. The method can be applied to a computer device, which is deployed with a cascaded cell detection model and a whole-slide classification model, including:

[0045] 101. Obtain a plurality of field images of a target digital slide.

[0046] The cell detection model is configured to detect cell images corresponding to positive cells in the field images and positive probabilities, and determine a slide category of the target digital slide for the first time according to the positive probabilities corresponding to the positive cells. The whole-slide classification model is configured to determine the slide category of the target digital slide for the second time if the target digital slide is determined as a suspicious positive slide for the first time. The slide category can be regarded as a pathological category of the target digital slide. The pathological category determined for the first time can include one of positive, negative, and suspicious positive.

[0047] The target digital slide can be a whole slide image (WSI).

[0048] In the embodiment, the step 101 can include: obtaining a target digital slide; performing foreground extraction based on the target digital slide to obtain a foreground region of the target digital slide; and cutting the foreground region into a plurality of field images of a preset size based on a grid division manner.

[0049] Specifically, a WSI can be cut into different field images by the grid division manner. Specifically, for a WSI, a foreground region (cell region) can be extracted by using a traditional image segmentation method, and then the foreground region is cut into field images of a fixed size based on the grid division manner. The fixed size of the field images is not limited in the present application, and can be determined according to the cell detection model. For example, the fixed size can be set to 1280x1280.

[0050] For example, referring to Fig. 1 and Fig. 2, Figure 2 The WSI is input into the computer device, the computer device extracts a foreground region (cell region) of the WSI by using a traditional image segmentation method, and then cuts the foreground region into field images of a fixed size based on the grid division manner. Then, the field images obtained by the cutting are input into the cell detection model.

[0051] It is easy to understand that the number of cells in a WSI is usually very large. In order to reduce the difficulty of subsequent cell feature extraction and improve the efficiency of subsequent cell processing, the WSI can be cut into a plurality of field images, and then each field image is processed subsequently.

[0052] 102, extract the cell image feature and the positive probability corresponding to the positive cell in each field of view image based on the cell detection model.

[0053] The cell detection model is trained according to the positive field of view image labeled with the positive cell and the negative field of view image without labeling the positive cell. By inputting the field of view image, the cell detection model can output the cell image feature and the positive probability corresponding to the positive cell in the field of view image.

[0054] Specifically, the cell detection model can adopt a commonly used two-stage detection model, respectively using Faster RCNN and ResNet50 pre-trained using ImageNet dataset, wherein the detection category of the model can be set to 1 category, that is, to detect the positive cell of the field of view image.

[0055] The positive probability is used to represent the probability that the cell is a positive cell. Specifically, if the positive probability is greater than a preset positive threshold, the pathological category of the positive cell can be determined as positive. If the positive probability is less than a preset negative threshold, the pathological category of the positive cell can be determined as negative. If the positive probability is not less than the preset negative threshold and not less than the preset negative threshold, the whole slide classification model cascaded with the cell detection model is needed to determine the pathological category of the field of view image. In this way, based on the cascading system of the two models, the accuracy of the model case analysis can be greatly improved.

[0056] 103, determine the target positive cell in the target digital slice based on the positive probability corresponding to the positive cell in the field of view image.

[0057] The positive probability corresponding to the target positive cell can be used to determine the slice category of the target digital slice for the first time. The first determined slice category can include one of the positive slice, the negative slice and the suspicious positive slice.

[0058] In some embodiments, step 103 can mainly include determining the positive cell with the maximum positive probability among all positive cells of the target digital slice as the target positive cell.

[0059] Specifically, the maximum positive probability in the positive probability is determined as the positive probability corresponding to the target positive cell. By matching the positive probability corresponding to the target positive cell with each probability interval, that is, by matching the maximum positive probability in the positive probability with each probability interval, if the maximum positive probability matches the positive probability interval, it can be determined that the field of view image has a cell with a positive pathological feature, and the slice category of the target digital slice can be determined as a positive slice. If the maximum positive probability matches the negative probability interval, it can be determined that the positive probability of all cells matches the negative probability interval, and the slice category of the target digital slice can be determined as a negative slice.

[0060] In some embodiments, step 103 can mainly include: ranking the positive cells in descending order of the positive probability; determining the positive cells located in the front first preset number of the ranking order as the target positive cells in the target digital slice; and averaging the positive probability corresponding to the positive cells located in the front first preset number of the ranking order, and determining the average value as the positive probability corresponding to the target positive cells.

[0061] In the present application, the first preset number is not limited. Generally, the number of cells is of a large order of magnitude, and therefore, the positive cells located in the front first preset number of the ranking order can be taken as the target positive cells. Specifically, the target positive cells can be considered as the cells with more prominent positive features among all the positive cells. Therefore, the positive probability corresponding to the first preset number of positive cells can be averaged, that is, the positive probability corresponding to the positive cells with more prominent positive features is averaged, and the average value is determined as the positive probability corresponding to the target positive cells, and then the slice category of the target digital slice is determined according to the positive probability corresponding to the target positive cells. For example, the first three positive cells in the ranking order can be determined as the target positive cells, and the positive probability corresponding to the first three positive cells in the ranking order is averaged, and the average value is determined as the positive probability corresponding to the target positive cells. If the positive probability corresponding to the target positive cells matches the positive probability interval, it can be determined that the field of view image contains cells with positive pathological features, and it can be determined that the slice category of the target digital slice is positive. If the positive probability corresponding to the target positive cells matches the negative probability interval, it can be determined that the positive probability of all positive cells matches the negative probability interval, and it can be determined that the slice category of the target digital slice is negative.

[0062] In the present embodiment, after the cell detection model obtains the detection results of all the field of view images, the detection results can include the cell image features and the positive probability corresponding to the positive cells in all the field of view images. The detection results of all the field of view images can be combined, and all the positive cells can be ranked based on the combined detection results according to the positive probability. The positive cells located in the front certain number of the ranking order are taken, and the positive cell with the largest probability among the first certain number of positive cells is determined as the target positive cell.

[0063] Similarly, in the present embodiment, after the cell detection model obtains the detection results of all field of view images, it can merge the detection results of all field of view images, and sort the positive cells according to the positive probability based on the merged detection results, take the first certain number of positive cells in the sorting order, and then determine the first first preset number of positive cells in the first certain number of positive cells as target positive cells, and average the positive probability corresponding to the first first preset number of positive cells, and determine the average value as the positive probability corresponding to the target positive cell.

[0064] Wherein, the first certain number of positive cells can be the following first second preset number of positive cells. That is, the first second preset number of positive cells can be taken from the merged detection results as suspicious positive cells, and then the first first preset number of suspicious positive cells in the first second preset number of suspicious positive cells are determined as target positive cells, and the positive probability corresponding to the target positive cell is determined according to the positive probability corresponding to the first first preset number of suspicious positive cells in the first second preset number of suspicious positive cells.

[0065] In the present embodiment, the diagnosis scene of cervical liquid-based cells is taken as an example. Because the diagnosis of cervical liquid-based cells is mainly for screening intraepithelial lesions, the positive cells can be but are not limited to at least 6 kinds of positive cells, which are as follows: atypical squamous cells of undetermined significance (ASC-US), low squamous intraepithelial lesion (LSIL), atypical squamous cell-cannot exclude HISL (ASCH), high squamous intraepithelial lesion (HSIL), squamous cell carcinoma (SCC) and atypical glandular cells (AGC). Among them, the difference between the above positive cells can be but is not limited to being in different pathological periods, for example, atypical squamous cells tend to be high-grade lesions, and are in a high-grade lesion period.

[0066] In some embodiments, the method can further include: if the positive probability corresponding to the target positive cell matches the positive probability interval, determining that the slice category of the target digital slice is a positive slice, wherein the minimum probability in the positive probability interval is greater than the maximum probability in the suspicious probability interval; and if the positive probability corresponding to the target positive cell matches the negative probability interval, determining that the slice category of the target digital slice is a negative slice, wherein the maximum probability in the negative probability interval is less than the minimum probability in the suspicious probability interval.

[0067] wherein the suspicious probability interval is an interval between the positive probability interval and the negative probability interval, for example, the positive probability interval is greater than t2, and the negative probability interval is less than t1, wherein t2 is greater than t1, and t1 to t2 can be considered as the suspicious probability interval. If the positive probability corresponding to the target positive cell is greater than t2, i.e., matches the positive probability interval, it can be determined that the pathology category of the field of view image is positive. If the positive probability corresponding to the target positive cell is less than t1, i.e., matches the negative probability interval, it can be determined that the pathology category of the field of view image is negative.

[0068] Specifically, if the positive probability corresponding to the target positive cell matches the suspicious probability interval, the target digital slice can be considered as a suspicious positive slice, i.e., it is currently unable to accurately determine whether the target digital slice is a positive slice or a negative slice. In order to accurately determine the slice category of the target digital slice, the slice positive probability of the target digital slice can be accurately determined based on the cell image features and the positive probability corresponding to the positive cell according to the full-slice classification model.

[0069] 104. When the positive probability corresponding to the target positive cell matches the suspicious probability interval, the slice category of the target digital slice is determined based on the cell image features and the positive probability corresponding to the positive cell according to the full-slice classification model.

[0070] wherein when the positive probability corresponding to the target positive cell matches the suspicious probability interval, the slice category of the target digital slice can be determined for the first time as a suspicious positive slice, and then the second determination is performed based on the cell image features and the positive probability corresponding to the positive cell according to the full-slice classification model. The slice category determined in the second determination includes one of a positive slice and a negative slice, and the suspicious probability interval is an interval between the positive probability interval and the negative probability interval.

[0071] In the present embodiment, step 104 can mainly include: determining the slice positive probability corresponding to the target digital slice according to the cell image features and the positive probability corresponding to the positive cell; and determining the slice category of the target digital slice according to the slice positive probability.

[0072] The full-slide classification model is trained according to the cell image features corresponding to the suspicious positive cells of the WSI labeled with the slide category (positive slide or negative slide) output by the cell detection model. By inputting the cell image features corresponding to the suspicious positive cells, the full-slide classification model can obtain the slide category of the target digital slide.

[0073] The slide positive probability is used to represent the probability that the slide category of the target digital slide is a positive slide. According to the slide positive probability, the slide category of the target digital slide can be predicted. For example, if the slide positive probability of the target digital slide is greater than a preset full-slide positive threshold, it is determined that the slide category of the target digital slide is a positive slide; if the slide positive probability of the target digital slide is not greater than the preset full-slide positive threshold, it is determined that the slide category of the target digital slide is a negative slide.

[0074] In this embodiment, the step of "determining the slide positive probability corresponding to the target digital slide according to the cell image features corresponding to the positive cells and the positive probability" can mainly include: sorting the positive cells in descending order according to the positive probability, and determining the first second preset number of positive cells in the sorting order as suspicious positive cells; determining the slide positive probability corresponding to the target digital slide according to the cell image features corresponding to the suspicious positive cells and the cell image features corresponding to the target positive cells.

[0075] When the target positive cell is the first preset number of positive cells in descending order according to the positive probability, the feature corresponding to the target positive cell can be the fusion feature of the feature corresponding to the target positive cell.

[0076] Specifically, referring to Figure 2 , after the field-of-view image is input into the cell detection model, the cell detection model outputs the positive feature map corresponding to the positive cells in the field-of-view image and the positive probability. According to the sorting of the positive probability, the first second preset number of positive cell feature maps in the sorting can be selected as suspicious positive cell feature maps. In this application, the second preset number is not limited. For example, in this embodiment, the second preset number can be set to 32, that is, the first 32 positive cell feature maps in the sorting are selected as suspicious positive cell feature maps, and then the suspicious positive cell feature maps are input into the full-slide model.

[0077] In this embodiment, the step of "determining the slide positive probability corresponding to the target digital slice according to the cell image features corresponding to the suspected positive cells and the cell image features corresponding to the target positive cells" can include: fusing the cell image features corresponding to the suspected positive cells to obtain first fused features; fusing the first fused features and the cell image features corresponding to the target positive cells to obtain second fused features; and determining the slide positive probability corresponding to the target digital slice according to the second fused features.

[0078] Specifically, referring to Figure 3 , by inputting the cell image features corresponding to 32 suspected positive cells of the WSI into the whole-slide classification model, the self-attention module of the whole-slide classification model fuses the cell image features corresponding to the suspected positive cells, and all cell features are fused through the attention mechanism to obtain first fused features. Then, the whole-slide classification model further fuses the first fused features and the cell image features corresponding to the target positive cells for enhancement. The specific fusion manner is to concatenate the first fused features and the cell image features corresponding to the target positive cells, and then further fuse them through a linear layer to obtain second fused features. Then, the slide positive probability can be obtained through the classifier of the whole-slide classification model.

[0079] Specifically, by fusing the features of 32 suspected positive cells, the integrity of the features corresponding to the WSI can be enhanced, and the prediction accuracy of the slide positive probability of the WSI according to the fused features can be improved.

[0080] In some embodiments, before 101, it can also include: obtaining a sample image set, the sample image set including a plurality of labeled sample digital slice images and a true value label of each labeled sample digital slice image, the true value label including a labeled position of a sample positive cell on each labeled sample digital slice image and a true value positive probability of the sample positive cell on each sample digital slice image; and inputting the sample image set into an initial cell detection model to obtain the cell detection model through training.

[0081] It is easy to understand that the biggest feature of artificial intelligence is strong learning ability. By inputting the labeled sample data into the initial model with random parameters for training, adjusting the parameters of the initial model when errors occur, and through a large amount of training, the required model can be formed.

[0082] In some embodiments, the step of "inputting the sample image set into the initial cell detection model to train a cell detection model" can mainly include: inputting the sample image set into the initial cell detection model, and obtaining the predicted positions of sample positive cells and the predicted positive probabilities of the sample positive cells determined by the initial cell detection model according to the labeled sample digital slice images; determining a first loss function according to the predicted positions, the predicted positive probabilities, the labeled positions, and the true positive probabilities; training the initial cell detection model according to the first loss function to obtain the cell detection model.

[0083] In the present embodiment, the true value label of the labeled sample digital slice image includes the labeled positions of sample positive cells and the true positive probabilities of the sample positive cells. Specifically, the sample positive cells can be labeled by a labeled box, which mainly includes the center point coordinate information of the labeled box, the length and width information of the labeled box. Specifically, the sample image set can also include unlabeled sample digital slice images, i.e., negative samples. The labeled sample digital slice images can be referred to as positive samples, and the unlabeled sample digital slice images can be referred to as negative samples, which can be distributed in proportion. Then, the sample image set is input into the initial cell detection model, and the loss function is determined according to the results output by the initial cell detection model and the true value label of the labeled sample digital slice image, and the initial cell detection model is trained.

[0084] In the present embodiment, the step of "determining a first loss function according to the predicted positions, the predicted positive probabilities, the labeled positions, and the true positive probabilities" can include: determining first verification samples according to a first preset proportion and the labeled sample digital slice images; determining the first loss function according to the predicted positions of sample positive cells in the first verification samples, the predicted positive probabilities of the sample positive cells in the first verification samples, the labeled positions of the sample positive cells in the first verification samples, and the true positive probabilities of the sample positive cells in the first verification samples determined by the initial cell detection model according to the first verification samples.

[0085] It is easy to understand that, in order to improve the training efficiency, a part of sample digital slice images can be selected from the sample image set as verification samples according to a proportion, and the loss function is determined according to the verification samples, and the initial detection model is trained. It is worth noting that the first preset proportion is not limited by the present application and can be customized.

[0086] For example, please refer to Figure 4 wherein the sample digital slice image on the left with a labeled box is a positive sample, and the sample digital slice image on the right without a labeled box is a negative sample.

[0087] In the embodiment, the method can further include: extracting cell image features corresponding to the suspicious positive cells in the plurality of sample digital slice images based on the trained cell detection model; and inputting the cell image features corresponding to the suspicious positive cells into the initial whole-slide classification model to train the whole-slide classification model.

[0088] Specifically, the true value label of the labeled sample digital slice image further includes a true value slice positive probability of the sample digital slice image. The features of the suspicious positive cells are input into the initial whole-slide classification model, and a loss function is determined according to a result output by the initial whole-slide classification model and the true value slice positive probability corresponding to the sample digital slice image, so as to train the initial whole-slide classification model to obtain the whole-slide classification model.

[0089] In some embodiments, the step of "inputting the cell image features corresponding to the suspicious positive cells into the initial whole-slide classification model to train the whole-slide classification model" can include: inputting the cell image features corresponding to the suspicious positive cells into the initial whole-slide classification model, and obtaining a predicted slice positive probability of the labeled sample digital slice image determined by the initial whole-slide classification model according to the cell image features corresponding to the suspicious positive cells; determining a second loss function according to the predicted slice positive probability and the true value slice positive probability of the labeled sample digital slice image; and training the initial whole-slide classification model according to the second loss function to obtain the whole-slide classification model.

[0090] Similarly, in order to improve the training efficiency, a part of the sample digital slice images can be selected from the sample image set as verification samples according to a proportion, and the second loss function is determined according to the verification samples to train the initial whole-slide classification model. The step of "determining a second loss function according to the predicted slice positive probability and the true value slice positive probability of the labeled sample digital slice image" can include: determining second verification samples according to the second preset proportion and the labeled sample digital slice images; and determining the second loss function according to a predicted slice positive probability of the second verification samples determined by the whole-slide classification model according to the plurality of suspicious positive cells of the second verification samples and a true value slice positive probability of the second verification samples.

[0091] Specifically, the whole-slide classification model trained can output a slice positive probability of a WSI according to features of a plurality of suspicious positive cells of the WSI. The slice positive probability is used to represent a positive probability of the slice of the WSI.

[0092] In order to better illustrate the image processing method provided by the embodiments of the present application, please refer to Figure 5 The flow of the image processing method provided by the embodiments of the present application can be summarized as follows:

[0093] Step 201, obtaining a plurality of field of view images of a target digital slice through a cell detection model.

[0094] The target digital slice can be a whole slide image (WSI). The WSI can be divided into different field of view images through a grid division method. Specifically, for a WSI, first, a traditional image segmentation method can be used to extract a foreground region (a cell region), and then the foreground region is divided into field of view images of a fixed size based on a grid division method. The fixed size of the field of view images is not limited in the present application and can be determined according to the cell detection model. For example, the fixed size can be set to 1280x1280.

[0095] Step 202, extracting cell image features and a positive probability corresponding to a positive cell in each field of view image based on the cell detection model.

[0096] The cell detection model is trained according to positive field of view images labeled with positive cells and negative field of view images without labeled positive cells. By inputting the field of view images, the cell detection model can output the cell image features and the positive probability corresponding to the positive cells in the field of view images.

[0097] Step 203, determining a target positive cell in the target digital slice and a positive probability corresponding to the target positive cell based on the positive probability corresponding to the positive cells in the field of view images.

[0098] In some embodiments, step 203 can mainly include determining a positive cell with the maximum positive probability among all positive cells in the target digital slice as the target positive cell.

[0099] Specifically, the maximum positive probability in the positive probability is determined as the positive probability corresponding to the target positive cell. By matching the maximum positive probability with each probability interval, if the maximum positive probability matches a positive probability interval, it can be determined that the field of view image contains a cell with a positive pathological feature, and it can be determined that the slice category of the target digital slice is a positive slice. If the maximum positive probability matches a negative probability interval, it can be determined that the positive probability of all cells matches the negative probability interval, and it can be determined that the slice category of the target digital slice is a negative slice.

[0100] In some embodiments, step 103 can mainly include: sorting the positive cells in descending order of the positive probability; determining the first preset number of positive cells in the sorting order as the target positive cells in the target digital slice; and averaging the positive probabilities corresponding to the first preset number of positive cells in the sorting order, and determining the average value as the positive probability corresponding to the target positive cell.

[0101] Wherein, the first preset number is not limited in the present application. Generally, the number of cells is large in order of magnitude, therefore, the first preset number of positive cells in the sorting order can be taken as the target positive cells. Specifically, the target positive cells can be considered as the cells with more significant positive features among all positive cells. Therefore, the average of the positive probabilities corresponding to the first preset number of positive cells can be calculated, that is, the average of the positive probabilities of the positive cells with more significant positive features, and the average value is determined as the positive probability corresponding to the target positive cells, and then the slice category of the target digital slice is determined according to the positive probability corresponding to the target positive cells. For example, the first three positive cells in the sorting order can be determined as the target positive cells, and the average of the positive probabilities corresponding to the first three positive cells in the sorting order is calculated, and the average value is determined as the positive probability corresponding to the target positive cells, if the positive probability corresponding to the target positive cells matches the positive probability interval, it can be determined that the field image exists cells with positive pathological features, and it can be determined that the slice category of the target digital slice is positive. If the positive probability corresponding to the target positive cells matches the negative probability interval, it can be determined that the positive probability of all positive cells matches the negative probability interval, and it can be determined that the slice category of the target digital slice is negative.

[0102] Step 204, the slice category of the target digital slice is determined for the first time according to the positive probability corresponding to the target positive cells.

[0103] Wherein, the slice category determined for the first time can include one of the positive slice, the negative slice, and the suspicious positive slice.

[0104] Specifically, if the positive probability corresponding to the target positive cells matches the positive probability interval, it is determined that the target digital slice is a positive slice, wherein the minimum probability in the positive probability interval is greater than the maximum probability in the suspicious probability interval. If the positive probability corresponding to the target positive cells matches the negative probability interval, it is determined that the target digital slice is a negative slice, wherein the maximum probability in the negative probability interval is less than the minimum probability in the suspicious probability interval.

[0105] Step 205, if the positive probability corresponding to the target positive cells matches the suspicious probability interval, it is determined that the slice category of the target digital slice is a suspicious positive slice, and based on the whole slice classification model, the slice positive probability of the target digital slice is determined according to the cell image features corresponding to the suspicious positive cells and the cell image features corresponding to the target positive cells, and the slice category of the target digital slice is determined for the second time according to the slice positive probability.

[0106] Wherein, the suspicious probability interval is a probability interval between the positive probability interval and the negative probability interval, and the suspicious positive cell is a positive cell located in the first second preset number of positive cells in the order of descending positive probability.

[0107] Wherein, the second determination of the slice category includes one of the positive slice and the negative slice.

[0108] Wherein, the whole slice classification model is trained according to the cell image features corresponding to the suspicious positive cells of the WSI labeled with the pathological category (positive or negative) output by the cell detection model, and the slice positive probability of the target digital slice can be determined by inputting the cell image features corresponding to the suspicious positive cells. Wherein, the slice positive probability is used to represent the probability that the target digital slice is a positive slice. According to the slice positive probability, the pathological category of the target digital slice can be predicted. For example, if the slice positive probability of the target digital slice is greater than the preset whole slice positive threshold, the target digital slice is determined to be a positive slice; if the slice positive probability of the target digital slice is not greater than the preset whole slice positive threshold, the target digital slice is determined to be a negative slice.

[0109] Wherein, the suspicious probability interval is a probability interval between the positive probability interval and the negative probability interval, for example, the positive probability interval is greater than t2, and the negative probability interval is less than t1, wherein t2 is greater than t1, then t1 to t2 can be considered as the suspicious probability interval. If the positive probability corresponding to the target positive cell is greater than t2, that is, it matches the positive probability interval, then the pathological category of the field image can be determined to be positive. If the positive probability corresponding to the target positive cell is less than t1, that is, it matches the negative probability interval, then the pathological category of the field image can be determined to be negative.

[0110] In this embodiment, the image processing method can be applied in the cervical liquid-based cell diagnosis scenario, for example, the cervical liquid-based cells are scanned into a WSI, then the WSI is divided into field images of a fixed size, and the field images are input into the cell detection model. The cell detection model extracts the cell image features and the positive probability corresponding to the positive cells in the field images according to the field images, and then if the positive probability corresponding to the target positive cell (the maximum positive probability in the positive probability) matches the suspicious probability interval, the cell image features corresponding to the suspicious positive cells are input into the whole-slide classification model. The whole-slide classification model determines the slide positive probability of the WSI according to the cell image features corresponding to the suspicious positive cells and the cell image features corresponding to the target positive cell, and determines the pathological category (positive slide or negative slide) of the WSI according to the slide positive probability. It is easy to understand that the image processing method can assist the pathologist in cervical liquid-based cell diagnosis, reduce the workload of the pathologist, improve the work efficiency of the pathologist, and can replace the pathologist to perform cervical liquid-based cell diagnosis to determine the pathological category of the pathological slide and record the corresponding positive cell region in the positive slide, thereby achieving the effect of improving the diagnosis efficiency of the cervical liquid-based cells.

[0111] All the above technical solutions can be combined to form optional embodiments of the present application, which will not be described one by one here.

[0112] In the embodiments of the present application, the image processing method is used to acquire a plurality of field images of a target digital slide, then based on the cell detection model, the cell image features and the positive probability corresponding to the positive cells in each field image are extracted, then based on the positive probability corresponding to the positive cells in the field images, the target positive cell in the target digital slide is determined, and when the positive probability corresponding to the target positive cell matches the suspicious probability interval, based on the whole-slide classification model, the slide category of the target digital slide is determined according to the cell image features and the positive probability corresponding to the positive cells. In the embodiments of the present application, the cell detection model and the whole-slide classification model are cascaded, the positive cells in the target digital slide are detected based on the cell detection model, the feature map and the positive probability corresponding to the positive cells are obtained, the target positive cell is determined first, and the slide category of the target digital slide is determined for the first time according to the positive probability corresponding to the target positive cell. If the positive probability corresponding to the target positive cell matches the suspicious probability interval, the slide category of the target digital slide is determined based on the whole-slide classification model, so as to determine the pathological category of the target digital slide, assist the pathologist in cytological diagnosis, reduce the workload of the doctor, and improve the efficiency and accuracy.

[0113] In order to better implement the image processing method of the embodiments of the present application, the embodiments of the present application also provide an image processing device. Please refer to Figure 6 ,Figure 6 A first structural schematic diagram of an image processing apparatus provided by an embodiment of the present application is shown in the figure. The image processing apparatus 10 can be applied to a computer device, and the computer device is deployed with a cascaded cell detection model and a whole-slide classification model, which comprises:

[0114] A first acquisition module 11 is configured to acquire a plurality of field-of-view images of a target digital slide;

[0115] A second acquisition module 12 is configured to extract cell image features and a positive probability corresponding to positive cells in each field-of-view image based on the cell detection model;

[0116] A first determination module 13 is configured to determine a target positive cell in the target digital slide based on the positive probability corresponding to the positive cells in the field-of-view images.

[0117] A second determination module 14 is configured to determine a slide category of the target digital slide based on the features and the positive probability corresponding to the positive cells by the whole-slide classification model when the positive probability corresponding to the target positive cell matches a suspicious probability interval.

[0118] Optionally, the first determination module 13 can be configured to sort the positive cells in descending order of the positive probability, determine the first preset number of positive cells in the front of the sorting order as the target positive cells in the target digital slide, and average the positive probability corresponding to the first preset number of positive cells in the front of the sorting order, and determine the average value as the positive probability corresponding to the target positive cells.

[0119] Optionally, the second determination module 14 can be configured to determine a slide positive probability corresponding to the target digital slide according to the cell image features and the positive probability corresponding to the positive cells, and determine the slide category of the target digital slide according to the slide positive probability.

[0120] Optionally, the second determination module 14 can be configured to sort the positive cells in descending order of the positive probability, and determine the second preset number of positive cells in the front of the sorting order as suspicious positive cells, and determine a slide positive probability of the target digital slide according to the cell image features corresponding to the suspicious positive cells and the cell image features corresponding to the target positive cells.

[0121] Optionally, the second determination module 14 can be configured to fuse the cell image features corresponding to the suspicious positive cells to obtain first fused features, fuse the first fused features and the cell image features corresponding to the target positive cells to obtain second fused features, and determine the slide positive probability of the target digital slide according to the second fused features.

[0122] Optionally, the image processing apparatus 10 can further include a third determination module, which can be configured to: if the positive probability corresponding to the target positive cell matches the positive probability interval, determine that the slice category of the target digital slice is a positive slice, wherein the minimum probability in the positive probability interval is greater than the maximum probability in the suspicious probability interval; and if the positive probability corresponding to the target positive cell matches the negative probability interval, determine that the slice category of the target digital slice is a negative slice, wherein the maximum probability in the negative probability interval is less than the minimum probability in the suspicious probability interval.

[0123] Optionally, the first acquisition module 11 can be configured to: acquire a target digital slice; perform foreground extraction based on the target digital slice to obtain a foreground region of the target digital slice; and divide the foreground region into a plurality of field-of-view images of a preset size based on a grid division manner.

[0124] Optionally, the image processing apparatus 10 can further include a training module, which can be configured to: acquire a sample image set, the sample image set including a plurality of labeled sample digital slice images and a true value label of each labeled sample digital slice image, the true value label including a labeled position of a sample positive cell on each labeled sample digital slice image and a true value positive probability of the sample positive cell on each sample digital slice image; and input the sample image set into an initial cell detection model to obtain the cell detection model through training.

[0125] Optionally, the training module can be specifically configured to: input the sample image set into the initial cell detection model, and acquire a predicted position of a sample positive cell and a predicted positive probability of the sample positive cell determined by the initial cell detection model based on the labeled digital slice image; determine a first loss function based on the predicted position, the predicted positive probability, the labeled position, and the true value positive probability; and train the initial cell detection model based on the first loss function to obtain the cell detection model.

[0126] Optionally, the training module can be further configured to: extract cell image features corresponding to a plurality of sample suspicious positive cells in the sample digital slice images based on the cell detection model obtained through training; and input the cell image features corresponding to the plurality of sample suspicious positive cells into an initial whole-slice classification model to obtain the whole-slice classification model through training.

[0127] Optionally, the labeled sample digital slice image is also labeled with a true value slice positive probability of the labeled sample digital slice image. The training module can be specifically configured to: input cell image features corresponding to suspicious positive cells of a plurality of samples into the initial whole slice classification model, and obtain a predicted slice positive probability of the labeled sample digital slice image determined by the initial whole slice classification model according to the cell image features corresponding to the suspicious positive cells of the plurality of samples; determine a second loss function according to the predicted slice positive probability and the true value slice positive probability of the labeled sample digital slice image; and train the initial whole slice classification model according to the second loss function to obtain the whole slice classification model.

[0128] It should be noted that the functions of the modules in the image processing apparatus 10 in the embodiments of the present application can correspond to the specific implementation manners in the above-mentioned method embodiments, which will not be described here.

[0129] The above-mentioned various units in the image processing apparatus 10 can be realized by software, hardware and combinations thereof in whole or in part. The above-mentioned various units can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to the above-mentioned various units.

[0130] The image processing apparatus 10 can be integrated in a terminal or a server with a storage and a processor installed to have computing capability, or the image processing apparatus 10 is the terminal or the server.

[0131] The image processing apparatus 10 provided by the embodiments of the present application is applied to a computer device, and the computer device is deployed with a cascaded cell detection model and a whole slice classification model. The first acquisition module 11 acquires a plurality of field of view images of a target digital slice. Then, the second acquisition module 12 extracts cell image features and positive probabilities corresponding to positive cells in each field of view image based on the cell detection model. Then, the first determination module 13 determines a target positive cell in the target digital slice based on the cell image features and the positive probabilities corresponding to the positive cells in the field of view image. When the positive probability corresponding to the target positive cell matches a suspicious probability interval, the second determination module 14 determines a slice category of the target digital slice based on the whole slice classification model according to the cell image features and the positive probabilities corresponding to the positive cells, thereby assisting pathologists in cytological diagnosis, reducing the workload of the doctors, and improving the efficiency and accuracy.

[0132] Optionally, the present application further provides a computer device including a memory and a processor, and the memory stores a computer program. When the processor executes the computer program, the steps in the above-mentioned method embodiments are implemented.

[0133] Figure 7A structural schematic diagram of a computer device is provided for an embodiment of the present application. The computer device can be a terminal or a server as shown in the structural schematic diagram. Figure 1 As shown in the structural schematic diagram, the computer device 20 can include a communication interface 21, a memory 22, a processor 23 and a communication bus 24. The communication interface 21, the memory 22 and the processor 23 can communicate with each other through the communication bus 24. The communication interface 21 is configured to perform data communication between the computer device 20 and an external device. The memory 22 can be configured to store software programs and modules. The processor 23 can execute the software programs and modules stored in the memory 22, for example, software programs corresponding to the operations in the foregoing method embodiments. Figure 7 As shown in the structural schematic diagram, the computer device 20 can include a communication interface 21, a memory 22, a processor 23 and a communication bus 24. The communication interface 21, the memory 22 and the processor 23 can communicate with each other through the communication bus 24. The communication interface 21 is configured to perform data communication between the computer device 20 and an external device. The memory 22 can be configured to store software programs and modules. The processor 23 can execute the software programs and modules stored in the memory 22, for example, software programs corresponding to the operations in the foregoing method embodiments.

[0134] Optionally, the processor 23 can invoke the software programs and modules stored in the memory 22 to perform the following operations: obtaining a plurality of field-of-view images of a target digital slice; extracting, based on a cell detection model, cell image features and a positive probability corresponding to a positive cell in each field-of-view image; determining a target positive cell in the target digital slice based on the positive probability corresponding to the positive cell in the field-of-view image; and when the positive probability corresponding to the target positive cell matches a suspicious probability interval, determining a slice category of the target digital slice based on a whole-slice classification model according to the cell image features and the positive probability corresponding to the positive cell.

[0135] The present application also provides a computer readable storage medium for storing a computer program. The computer readable storage medium can be applied to a computer device, and the computer program causes the computer device to perform the corresponding procedures in the image processing method in the embodiments of the present application. For brevity, details are not repeated here.

[0136] The present application also provides a computer program product, which includes computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device performs the corresponding procedures in the image processing method in the embodiments of the present application. For brevity, details are not repeated here.

[0137] The present application also provides a computer program, which includes computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device performs the corresponding procedures in the image processing method in the embodiments of the present application. For brevity, details are not repeated here.

[0138] It should be understood that the processor of the embodiments of the present application can be an integrated circuit chip with a processing capability of signals. In the implementation process, each step of the method embodiments described above can be completed by the integrated logic circuit of hardware in the processor or the instructions in the form of software. The processor described above can be a general processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, a discrete gate or transistor logic device, a discrete hardware component. The disclosed methods, steps and logic block diagrams in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor or the like. The steps of the method disclosed in conjunction with the embodiments of the present application can be directly embodied as a hardware code processor for execution, or can be executed by a combination of hardware and software modules in the code processor. The software module can be located in a random memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register or other mature storage medium in the art. The storage medium is located in the memory, and the processor reads the information in the memory, and combines the hardware to complete the steps of the above method.

[0139] It is to be understood that the memory in the embodiments of the present application can be a volatile memory or a nonvolatile memory, or can include both volatile and nonvolatile memory. Among them, the nonvolatile memory can be a read-only memory (Read-Only Memory, ROM), a programmable read-only memory (Programmable ROM, PROM), an erasable programmable read-only memory (Erasable PROM, EPROM), an electrically erasable programmable read-only memory (Electrically EPROM, EEPROM) or a flash memory. The volatile memory can be a random access memory (Random Access Memory, RAM) used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (Static RAM, SRAM), dynamic random access memory (Dynamic RAM, DRAM), synchronous dynamic random access memory (Synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (Double Data Rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (Enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (Synchlink DRAM, SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DR RAM). It should be noted that the memory of the system and method described herein is intended to include, but not limited to, these and any other suitable types of memory.

[0140] It should be understood that the above-mentioned memory is exemplary but not limiting, for example, the memory in the embodiments of the present application can also be static random access memory (static RAM, SRAM), dynamic random access memory (dynamic RAM, DRAM), synchronous dynamic random access memory (synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (synch link DRAM, SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DR RAM) and the like. That is, the memory in the embodiments of the present application is intended to include, but not limited to, these and any other suitable types of memory.

[0141] Those skilled in the art can clearly understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0142] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0143] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division, and there can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0144] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e. they can be located in one place or distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0145] In addition, each functional unit in the embodiments of the present application can be integrated into a processing unit, or each unit can exist physically, or two or more units can be integrated into one unit.

[0146] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the parts that make contributions to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server) to execute all or part of the steps of the methods described in the embodiments of the present application. The foregoing storage medium includes: U disk, mobile hard disk, ROM, RAM, magnetic disk or optical disk and various program codes that can be stored.

[0147] The above description is only specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An image processing method, characterized in that, Applied to a computer device, the computer device being deployed with a cascaded cell detection model and a whole-slice classification model, the method includes: Acquire multiple field-view images of the target digital slice; Based on the cell detection model, cell image features and positive probability corresponding to positive cells in each field of view image are extracted; Based on the positive probability corresponding to the positive cells in the visual field image, the target positive cells in the target digital slice are determined; When the positive probability corresponding to the target positive cell matches the suspicious probability interval, the slice category of the target digital slice is determined based on the whole slice classification model and the cell image features and positive probability corresponding to the positive cell.

2. The image processing method according to claim 1, characterized in that, The determination of target positive cells in the target digital slice based on the positive probability corresponding to positive cells in the visual field image includes: The positive cells are sorted in descending order according to their positive probability. The positive cells that are among the first predetermined number in the sorting order are identified as the target positive cells in the target digital slice; and The average value of the positive probabilities corresponding to the first preset number of positive cells in the sorting order is calculated, and the average value is determined as the positive probability corresponding to the target positive cell.

3. The image processing method as described in claim 1, characterized in that, The step of determining the slice category of the target digital slice based on the whole-slice classification model, according to the cell image features and positive probability corresponding to the positive cells, includes: Based on the cell image features and positive probability corresponding to the positive cells, the positive probability of the slice corresponding to the target digital slice is determined; The slice category of the target digital slice is determined based on the positive probability of the slice.

4. The image processing method as described in claim 3, characterized in that, The step of determining the positive probability of the target digital slice based on the cell image features and positive probability corresponding to the positive cells includes: The positive cells are sorted in descending order according to the positive probability, and the positive cells in the first second preset number of the sorted order are identified as suspected positive cells. Based on the cell image features corresponding to the suspected positive cells and the cell image features corresponding to the target positive cells, the positive probability of the slice corresponding to the target digital slice is determined.

5. The image processing method as described in claim 4, characterized in that, The step of determining the positive probability of the target digital slice based on the cell image features corresponding to the suspected positive cells and the cell image features corresponding to the target positive cells includes: The cell image features corresponding to the suspected positive cells are fused to obtain the first fused feature; The first fusion feature is fused with the cell image feature corresponding to the target positive cell to obtain the second fusion feature; The positive probability of the target digital slice is determined based on the second fusion feature.

6. The image processing method as described in claim 1, characterized in that, The method further includes: If the positive probability corresponding to the target positive cell matches the positive probability interval, then the slice category of the target digital slice is determined to be a positive slice, wherein the minimum probability within the positive probability interval is greater than the maximum probability within the suspicious probability interval; If the positive probability corresponding to the target positive cell matches the negative probability interval, then the slice category of the target digital slice is determined to be a negative slice, wherein the maximum probability within the negative probability interval is less than the minimum probability within the suspicious probability interval.

7. The image processing method as described in claim 1, characterized in that, The acquisition of multiple field-view images of the target digital slice includes: Obtain the target number slice; Foreground extraction is performed based on the target digital slice to obtain the foreground region of the target digital slice; The foreground region is divided into multiple field-view images of a preset size based on a grid division method.

8. The image processing method according to any one of claims 1-7, characterized in that, Before acquiring multiple field-of-view images of the target digital slice, the process includes: Obtain a set of sample images, which includes multiple labeled digital slice images and a ground truth label for each labeled digital slice image. The ground truth label includes the labeled position of a positive cell on each labeled digital slice image and the ground truth positive probability of the positive cell on each labeled digital slice image. The sample image set is input into the initial cell detection model to train the cell detection model.

9. The image processing method as described in claim 8, characterized in that, The step of inputting the sample image set into the initial cell detection model to train the cell detection model includes: The sample image set is input into the initial cell detection model, and the predicted location of the positive cells in the sample determined by the initial cell detection model based on the labeled sample digital slice image, as well as the predicted positive probability of the positive cells in the sample, are obtained. A first loss function is determined based on the predicted location, the predicted positive probability, the labeled location, and the true positive probability. The initial cell detection model is trained based on the first loss function to obtain the cell detection model.

10. The image processing method as described in claim 8, characterized in that, The method further includes: Based on the cell detection model obtained through training, cell image features corresponding to suspected positive cells in the multiple sample digital slice images are extracted; The cell image features corresponding to the multiple suspected positive cells in the samples are input into the initial whole-slice classification model to train the whole-slice classification model.

11. The image processing method as described in claim 10, characterized in that, The ground truth label of the labeled sample digital slice image also includes the ground truth slice positive probability of the labeled sample digital slice image. The step of inputting the cell image features corresponding to the multiple suspected positive cells into the initial whole-slice classification model to train the whole-slice classification model includes: The cell image features corresponding to the multiple suspected positive cells in the samples are input into the initial whole-slice classification model, and the predicted slice positive probability of the labeled sample digital slice image is obtained by the initial whole-slice classification model based on the cell image features corresponding to the multiple suspected positive cells in the samples. The second loss function is determined based on the predicted slice positive probability and the ground truth slice positive probability of the labeled sample digital slice image; The initial whole-piece classification model is trained according to the second loss function to obtain the whole-piece classification model.

12. An image processing apparatus, characterized in that, Applied to computer equipment, the computer equipment being deployed with cascaded cell detection models and whole-slice classification models, including: The first acquisition module is used to acquire multiple field-view images of the target digital slice; The second acquisition module is used to extract cell image features and positive probability corresponding to positive cells in each field of view image based on the cell detection model. The first determining module is used to determine the target positive cells in the target digital slice based on the positive probability corresponding to the positive cells in the visual field image; The second determining module is used to determine the target digital slice category based on the whole slice classification model, according to the cell image features and positive probability corresponding to the positive cell, when the positive probability corresponding to the target positive cell matches the suspicious probability interval.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted for loading by a processor to perform the steps of the image processing method as described in any one of claims 1-11.

14. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing a computer program, and the processor executing the steps of the image processing method according to any one of claims 1-11 by calling the computer program stored in the memory.

15. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the image processing method according to any one of claims 1-11.

Images