Blood cell number counting method, system and equipment based on image recognition and medium

Abstract

The invention discloses a blood cell number counting method, system and device based on image recognition and a medium, and relates to the technical field of image data processing, and the method comprises the steps: recognizing a blood smear image based on a recognition model, and judging and marking the cell type and the accuracy probability of each cell image; generating a probability distribution diagram based on the position distribution rule of each type of blood cells on the blood smear; obtaining the cell type, the accurate probability and the position coordinate of each cell image, giving a reliability weight to the accurate probability in combination with the probability distribution diagram, generating a type judgment reliability value of each cell image, comparing the type judgment reliability value with a set threshold value, determining the cell type according to a comparison result, marking the cell type, and counting and outputting the quantity of each type of blood cells according to the marking. According to the scheme, while the blood cell image is identified based on the AI, the blood cell morphology and the number of the cell overlapping image are calculated in combination with the probability distribution, the accuracy of blood cell morphology identification and counting is improved, and reliable data reference is provided for decision making of doctors.

Description

Technical Field

[0001] The present application relates to the technical field of image data processing, and in particular to a method, system, device and medium for counting the number of blood cells based on image recognition. Background Art

[0002] In the field of medical testing, blood cell morphology examination is crucial for the diagnosis of many diseases, such as leukemia, anemia, infection, etc. Traditional blood cell morphology diagnosis mainly relies on inspectors to manually observe blood smears through a microscope, identify the morphology, quantity and abnormal characteristics of various blood cells based on experience, and then write a diagnosis report.

[0003] Currently, more and more medical testing institutions and medical equipment manufacturers have introduced AI image analysis technology into the field of blood cell morphology diagnosis. They use AI analysis modules with built-in blood cell recognition models to automatically recognize blood cell images in blood smears. They can accurately identify various blood cells such as red blood cells, white blood cells, platelets and their subtypes. At the same time, they can accurately detect cell morphological abnormalities, such as uneven size, abnormal shape, nuclear and cytoplasmic changes, etc., and output analysis results containing detailed information such as cell type, quantity, abnormal characteristics, etc. Then, based on the correspondence between blood cell morphological changes and common diseases, they generate auxiliary diagnosis reports for doctors to refer to and make decisions.

[0004] As more and more blood cell image materials are fed into the training model, the recognition accuracy of the blood cell recognition model has reached more than 95%. The main factor currently affecting the identification and counting of blood cell morphology is that the cells on the blood smear will overlap, and the overlapping blood cell morphology is difficult to be effectively identified by the recognition model. In fact, when different types of blood cells completely overlap, even manual microscopy cannot accurately judge and distinguish the cell morphology and quantity that constitute the overlapping image, and manual microscopy is inefficient, which has an adverse effect on the accurate identification and counting of blood cell morphology. Summary of the invention

[0005] In order to solve the problem of insufficient accuracy and low efficiency of existing blood cell morphology diagnosis, the first purpose of this application is to provide a blood cell counting method based on image recognition, which identifies blood cell images based on artificial intelligence technology, and at the same time combines probability distribution to infer the blood cell morphology and quantity of cell overlapping images, further improving the accuracy and efficiency of blood cell morphology identification and counting, and providing reliable data reference for doctors' decision-making. In order to realize the above-mentioned blood cell counting method based on image recognition, the second purpose of this application is to provide a blood cell counting system based on image recognition, which is not only optimized in terms of recognition software, but also in terms of hardware configuration, so as to further improve the accuracy of blood cell counting. The third purpose is to provide a blood cell counting device based on image recognition, and finally it is proposed to protect a computer-readable storage medium, on which a computer program module for implementing the blood cell counting method based on image recognition is loaded. The specific scheme is as follows:

[0006] A blood cell counting method based on image recognition, comprising:

[0007] Identify blood smear images based on the recognition model, determine and annotate the cell type and accurate probability corresponding to each cell image;

[0008] Obtaining and generating a probability distribution diagram for characterizing the probability of each type of blood cell appearing at different positions on the blood smear based on the position distribution rules of different types of blood cells on the blood smear;

[0009] Obtaining the cell type and accurate probability corresponding to each cell image after determination, as well as the position coordinates of each cell image, assigning a reliability weight to the accurate probability in combination with the probability distribution graph, and generating a type determination reliability value corresponding to each cell image;

[0010] Compare the type determination reliability value corresponding to each cell image with a set threshold:

[0011] If the type determination reliability value corresponding to the cell image exceeds the set threshold, the sample is determined to be confirmed and its cell type is marked;

[0012] If the type determination reliability value corresponding to the cell image does not exceed the set threshold, it is determined as a sample to be determined, and the cell type and quantity of the sample to be determined are confirmed and marked by manual and / or extended determination steps;

[0013] Based on the annotation information of each cell image in the blood smear image, the number of each type of blood cells is counted and output.

[0014] Through the above technical solution, the blood smear image is first identified based on artificial intelligence technology to determine the cell type corresponding to each cell image, and then the accuracy of the determination is evaluated, and the corresponding accurate probability is generated by association. Then, according to the distribution pattern of each type of cell on the blood smear, the reliability weight is assigned to the above accurate probability, thereby making the result of cell type determination more accurate. For some cell images with low determination accuracy, the accuracy of the determination result can be improved through weighted calculation, and finally the accuracy and efficiency of blood cell morphology identification and counting can be improved, providing reliable data reference for doctors' diagnosis and treatment decisions.

[0015] Furthermore, the cell image includes a single cell image and a cell overlapping image;

[0016] The recognition model stores the association relationship between a single cell image and a cell type, and between a cell overlapping image and a combination of cell types and quantities constituting the cell overlapping image, which is stored as a first relationship database;

[0017] In the blood cell counting method, the cell type corresponding to each cell image is determined to be multiple and each generated cell type is determined to correspond to an accurate probability;

[0018] Determine and label the cell type and accurate probability corresponding to each cell image, including:

[0019] Obtain a single cell image and a cell overlapping image in a blood smear image, search for matching cell images similar to the single cell image and the cell overlapping image based on the first relational database, obtain the corresponding cell type or cell type and quantity combination, and mark the accuracy probability for each matching result according to the similarity between the cell images.

[0020] Through the above technical solution, each cell image in the blood smear image corresponds to a cell category, and each determined cell category corresponds to an accurate probability, which is convenient for subsequent weighted calculation; at the same time, the cell overlapping image also corresponds to the cell type and quantity combination, providing a primary determination basis for the subsequent extended determination step.

[0021] Furthermore, the blood cell counting method further comprises:

[0022] Establishing the correlation between the overlapping probability of each type of cells, the overlapping position of cells and the blood smear making parameters, and storing them as a second relational database;

[0023] The extension determination step comprises:

[0024] Acquire overlapping cell images, acquire multiple corresponding cell type and quantity combinations through a first relational database, and configure a first accurate probability value for each combination;

[0025] Acquire the position coordinates of the cell overlapping images, configure a second accurate probability value for each of the cell type and quantity combinations according to the second relational database and the probability distribution map, and configure a third accurate probability value for the cell type in the combination;

[0026] Acquire blood smear making parameter data corresponding to the current blood smear image, and configure a fourth accurate probability value for each of the cell type and quantity combinations according to the second relational database;

[0027] Assigning reliable weight values ​​to the first accurate probability value, the second accurate probability value, the third accurate probability value and the fourth accurate probability value, and obtaining reliable values ​​for determining the cell type, quantity and class amount corresponding to each cell overlapping image according to a set algorithm;

[0028] Compare the above-mentioned reliable value of class quantity determination with a set threshold:

[0029] If the reliable value of the class quantity judgment corresponding to the cell overlapping image exceeds the set threshold, it is determined as a confirmed sample and its corresponding cell type and quantity are marked;

[0030] If the reliable value of the class quantity judgment corresponding to the cell overlapping image does not exceed the set threshold, it is judged as a sample to be determined, and the cell type and quantity of the sample to be determined are manually confirmed and marked;

[0031] The blood smear production parameters include the angle between the pusher and the slide, the moving speed of the pusher and the blood volume.

[0032] Through the above technical solution, the factors affecting the probability of cell overlap and the position of cell overlap are introduced into the determination of the cell type of the sample to be determined, which can effectively improve the accuracy of the determination of the cell type corresponding to the sample to be determined, and provide reliable data support for the doctor's subsequent diagnosis.

[0033] Furthermore, the blood cell counting method further comprises:

[0034] Dividing the whole or part of the blood smear image into a plurality of continuous statistical regions with equal areas;

[0035] Count the number of each type of cells contained in the determined sample in each statistical area, and calculate the proportion of each type of cells in the statistical area;

[0036] According to the ratio of the number of set type of cells in each statistical area, a trend curve of the ratio changing with the statistical area is generated by fitting;

[0037] Obtain cell types and quantity combinations corresponding to overlapping images of cells in the statistical area, and decompose the above cell types and quantity combinations into estimated quantities of cells of each type;

[0038] The estimated number of cells of each type is added to the statistical data of the current statistical area, and the ratio of the number of cells of each type in the current statistical area before and after the estimated number is added is calculated;

[0039] According to the degree of deviation between the above-mentioned quantity proportion ratio and the above-mentioned trend curve, a fifth accurate probability value is configured for each of the above-mentioned cell type and quantity combinations;

[0040] Assigning weight values ​​to the first accurate probability value, the second accurate probability value, the third accurate probability value, the fourth accurate probability value and the fifth accurate probability value, and calculating and obtaining the cell types and quantities contained in the cell overlapping image according to a set algorithm;

[0041] The criterion for dividing the statistical area is that the sum of the differences in the number of samples to be determined in each adjacent statistical area is the largest.

[0042] The distribution of each type of blood cell on a blood smear usually changes evenly and continuously, that is, the ratio of the number of each type of cells in each statistical area should be a smooth curve. Through the above technical scheme, when the number of each type of cells obtained by reverse inference from the uncertain sample, that is, the cell overlapping image, is added to the existing number of each type of cells, if the ratio of the number of each type of cells changes too much, that is, the number ratio in the current statistical area deviates significantly from the trend curve, it can be inferred that the number of each type of cells obtained by reverse inference does not satisfy the statistical probability distribution law, and thus the reliability of the inference is greatly reduced, and vice versa. Through the above technical scheme, even if some cell images in the blood smear image are difficult to determine their original cell composition due to cell overlap, manual or AI image recognition can be used to assign corresponding reliability weight values ​​to various cell types and quantity combinations through probability distribution, thereby further improving the reliability of cell type and quantity statistics.

[0043] Furthermore, the cell number counting method further comprises:

[0044] Receive and respond to external image review requests, query the target cell type based on the annotation information of each cell image in the blood smear image, and enhance or fade the set cell image.

[0045] Through the above technical solution, inspectors can quickly review the number and morphology of blood cells of a set type, and can quickly obtain judgment conclusions for some uncertain samples, thereby improving the efficiency of the entire blood cell counting.

[0046] Furthermore, the identification of the blood smear image based on the recognition model also includes an image preprocessing step: performing noise reduction, contrast enhancement, and grayscale correction operations on the acquired blood smear image to remove noise interference generated during the image acquisition process and highlight the morphological details of blood cells, and then outputting the preprocessed blood smear image to the recognition model.

[0047] Through the above technical solution, the image features of the blood smear image can be extracted efficiently and accurately, which is convenient for improving the recognition accuracy of the subsequent recognition model and improving the reliability of the counting results.

[0048] A blood cell counting system based on image recognition, comprising:

[0049] A data storage unit configured to acquire and store the blood smear image data to be identified and a probability distribution diagram for characterizing the position distribution law of different types of blood cells on the blood smear;

[0050] An image recognition unit, with a built-in recognition model, configured to be data-connected to the data storage unit, retrieve the blood smear image data and recognize the blood smear image based on the recognition model, and determine and mark the cell type, accurate probability and position coordinates corresponding to each cell image therein;

[0051] A reliability evaluation unit is configured to be connected to the image recognition unit data, and is used to obtain the cell type and accuracy probability corresponding to each cell image after determination, as well as the position coordinates of each cell image, assign a reliability weight to the accuracy probability in combination with the probability distribution diagram, and generate a type determination reliability value corresponding to each cell image;

[0052] The counting generation unit has a built-in comparison output module, which is configured to be data-connected with the reliability evaluation unit, receive the type determination reliability value corresponding to each cell image and compare it with a set threshold value: if it exceeds the set threshold value, the corresponding cell image is determined as a determined sample and its cell type is marked; if it does not exceed the set threshold value, it is determined as a sample to be determined and the cell type and quantity of the above-mentioned sample to be determined are confirmed and marked by manual and / or extended determination steps; based on the marked information of each cell image in the blood smear image, the number of each type of blood cells is counted and output.

[0053] Through the above technical scheme, the reliability weight value can be assigned to the determination result of the sample according to the position distribution law of different types of blood cells on the blood smear, thereby increasing the reliability of the determination result and making the final statistical number of each type of cells more reliable, providing more reliable data support for doctors to diagnose the disease.

[0054] Furthermore, the data storage unit also stores: a first correlation between the probability of cell overlap of each type, the cell overlap position and the blood smear production parameters, and a second correlation between the cell overlap image and the cell type and quantity combination constituting the cell overlap image;

[0055] The blood cell number counting system also includes:

[0056] The extended data preprocessing unit is configured to be data-connected with the data storage unit and the image recognition unit, obtain the cell overlapping image and obtain a plurality of cell type and quantity combinations corresponding to the cell overlapping image according to the first association relationship, and configure a first accurate probability value for each combination; obtain the position coordinates of the cell overlapping image, configure a second accurate probability value for each of the cell type and quantity combinations according to the second association relationship, and configure a third accurate probability value for each cell type in the combination according to the probability distribution diagram; obtain the blood smear making parameter data corresponding to the current blood smear image, and configure a fourth accurate probability value for each of the cell type and quantity combinations according to the second association relationship;

[0057] The first extended judgment unit is configured to be data-connected with the extended data preprocessing unit, assign reliable weight values ​​to the first accurate probability value, the second accurate probability value, the third accurate probability value and the fourth accurate probability value, and obtain the cell type, quantity and class quantity judgment reliability value corresponding to the cell overlapping image according to the set algorithm; compare the above-mentioned class quantity judgment reliability value with a set threshold value, and determine the cell type and quantity corresponding to the above-mentioned cell overlapping image according to the comparison result, or determine the cell type and quantity corresponding to the above-mentioned cell overlapping image by manual judgment.

[0058] Through the above technical solution, the blood smear production parameters that affect the cell distribution and overlap probability on the blood smear are incorporated into the determination probability analysis of uncertain samples, which can improve the reliability of the determination results.

[0059] Furthermore, the blood cell counting system also includes:

[0060] An image region segmentation unit configured to divide the whole or part of the blood smear image into a plurality of continuous statistical regions of equal area;

[0061] A quantity proportion statistical unit, configured to be data-connected to the count generating unit, for counting the quantity of each type of cells contained in the determined sample in each statistical area, and calculating the quantity proportion of each type of cells in the statistical area;

[0062] A reference curve generating unit is configured to generate a trend curve of the ratio changing with the statistical area according to the ratio of the number of set type of cells in each statistical area;

[0063] a disassembly and comparison unit, configured to be data-connected to the extended data preprocessing unit, and used to obtain the cell type and quantity combination corresponding to the overlapping image of cells in the statistical area, disassemble the above cell type and quantity combination into the estimated quantity of each type of cells, add the estimated quantity of each type of cells to the statistical data of the current statistical area, calculate the proportion of the quantity of each type of cells in the current statistical area before and after the above estimated quantity is added, and configure a fifth accurate probability value for each of the cell type and quantity combinations according to the degree of deviation of the above quantity proportion ratio from the trend curve;

[0064] The second extension determination unit is used to assign weight values ​​to the first accurate probability value, the second accurate probability value, the third accurate probability value, the fourth accurate probability value and the fifth accurate probability value, and calculate and obtain the cell types and quantities contained in the cell overlapping image according to a set algorithm.

[0065] Through the above technical solution, the reliability of blood cell counting results can be further improved.

[0066] A blood cell counting device based on image recognition, comprising:

[0067] An image acquisition device configured to acquire and output a blood smear image;

[0068] A blood smear making parameter acquisition device, configured to be data-connected to a smear machine that automatically makes a blood smear, for acquiring and outputting blood smear making parameter data;

[0069] The blood cell counting system based on image recognition as described above is configured to be data-connected with the image acquisition device and the blood smear making parameter acquisition device, and is used to acquire blood smear images and blood smear making parameter data, and output the number of each type of blood cells and annotation information after processing by the built-in program algorithm module; and

[0070] The information interaction device is configured to be data-connected to the blood cell counting system, and is used to receive the user's operation instructions and display and output the number of each type of blood cells and the marking information obtained from the system end.

[0071] A computer-readable storage medium is loaded with a program module for implementing the blood cell counting method based on image recognition as described above.

[0072] The above technical solution is helpful to promote the use of the method of the present application.

[0073] In summary, the present application includes at least one of the following beneficial technical effects:

[0074] (1) According to the position distribution law of different types of blood cells on the blood smear, the reliability weight value is assigned to the determination result of the sample to be determined, thereby increasing the reliability of the determination result, making the final statistical number of each type of cell more reliable, and providing more reliable data support for doctors to diagnose the disease;

[0075] (2) The factors that affect the probability of cell overlap and the position of cell overlap are introduced into the determination of the cell type of the sample to be determined, which effectively improves the accuracy of determining the cell type corresponding to the sample to be determined. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a schematic diagram of cell morphology on a blood smear;

[0077] Figure 2 It is a schematic diagram of the process of the blood cell counting method of the present application;

[0078] Figure 3 It is a probability distribution diagram of the probability of a set type of cell appearing at different locations on a blood smear;

[0079] Figure 4 It is a schematic diagram of a method for improving the reliability of blood cell count;

[0080] Figure 5 It is a functional module schematic diagram of the blood cell counting system of the present application;

[0081] Figure 6 It is a schematic diagram of a blood cell counting device of the present application.

[0082] Figure numerals: 1. data storage unit; 2. image recognition unit; 3. reliability assessment unit; 4. counting generation unit; 5. extension data preprocessing unit; 6. first extension determination unit; 7. image area segmentation unit; 8. quantity proportion statistics unit; 9. reference curve generation unit; 10. disassembly and comparison unit; 11. second extension determination unit; 100. image acquisition device; 200. blood smear production parameter acquisition device; 300. information interaction device; 400. blood cell number counting system. DETAILED DESCRIPTION

[0083] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings.

[0084] In the description of this specification, the description with reference to the terms "certain embodiments", "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiments or examples are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0085] Before describing the embodiments of the present application in detail, in order to more clearly illustrate the working principle of the embodiments of the present application, the relevant technologies or basic theories associated with blood cell morphology identification and counting are described here.

[0086] There are three main types of cells in the blood: red blood cells, white blood cells, and platelets (thrombus cells). Red blood cells are the most numerous cells in the blood, and their main function is to transport oxygen and carbon dioxide between the lungs and various functional tissues in the body. White blood cells are an important part of the immune system and there are multiple types, including neutrophils, lymphocytes, monocytes, and eosinophils, each with a corresponding morphology. Platelets are the smallest blood component and play a key role in hemostasis and coagulation.

[0087] Currently, most disease diagnosis and treatment processes require complete blood cell analysis. The main method is to make a blood smear and then observe the number and morphology of various types of cells under a microscope to use this as the basis for diagnosis of the disease.

[0088] The specific production process of the blood smear will not be repeated here. The current production of blood smears can be completed automatically by a smear machine. It should be pointed out here that different types of blood cells will show different distribution patterns on the blood smear. For example, the outermost layer of the blood smear is a large number of red blood cells, and there are some white blood cells and platelets in the middle area. White blood cells can be divided into granular and non-granular types according to their morphology. Among the granular white blood cells, neutrophils account for the largest proportion, followed by eosinophils and basophils. Among the non-granular white blood cells, monocytes have the largest volume, followed by lymphocytes, and platelets usually gather together to form small clusters. Different blood smear production parameters will form different cell distribution patterns on the blood smear. The above-mentioned blood smear production parameters mainly include: the volume of blood dripped on the slide, the angle between the pusher and the slide, and the speed of the pusher. In the automatic smear machine, the above parameters and even the temperature, the amount of dye and the pH value of the clean water can be accurately controlled.

[0089] In the field of cell image recognition, by feeding massive amounts of cell images accurately labeled with cell type information into deep learning algorithms, such as convolutional neural networks, we can train a recognition model that can accurately identify various types of cells. Currently, the recognition accuracy of the recognition model can reach more than 95%. As model training continues, the recognition accuracy of cell images will further increase in the future.

[0090] like Figure 1 The figure shows a blood smear image, from which we can see that each cell image is not completely independent and scattered, and many blood cells overlap. Therefore, it is difficult to accurately determine the type and number of blood cells when using the image recognition model for recognition. Figure 1 The basophils indicated by the middle arrows are stained darker, making it difficult to see the morphology and number of the overlapping cells behind them. Figure 1 The cells in the middle dotted box have obvious partial overlap.

[0091] Based on the introduction of the above-mentioned related technologies and basic theories, the specific contents of the implementation methods of this application are expanded below.

[0092] The present application embodiment discloses a method for counting the number of blood cells based on image recognition, such as Figure 2 As shown, it mainly includes the following steps:

[0093] S100, identifying the blood smear image based on the recognition model, determining and marking the cell type and accuracy probability corresponding to each cell image therein;

[0094] S200, obtaining and generating a probability distribution diagram for representing the probability of each type of blood cell appearing at different positions on the blood smear based on the position distribution rules of different types of blood cells on the blood smear;

[0095] S300, obtaining the cell type and accurate probability corresponding to each cell image after determination, as well as the position coordinates of each cell image, assigning a reliability weight to the accurate probability in combination with the probability distribution graph, and generating a type determination reliability value corresponding to each cell image;

[0096] S400, comparing the type determination reliability value corresponding to each cell image with a set threshold:

[0097] S410, if the type determination reliability value corresponding to the cell image exceeds the set threshold, it is determined to be a confirmed sample and its cell type is marked;

[0098] S420, if the type determination reliability value corresponding to the cell image does not exceed the set threshold, it is determined as a sample to be determined, and the cell type and quantity of the sample to be determined are confirmed and marked by manual and / or extended determination steps;

[0099] S500: Based on the annotation information of each cell image in the blood smear image, count and output the number of each type of blood cells.

[0100] In step S100, the recognition model uses an AI image analysis model, which has a built-in blood cell recognition model trained based on a deep learning algorithm. The above blood cell recognition model has been trained with a large amount of annotated blood cell image data, and can accurately identify various types of blood cells such as red blood cells, white blood cells, platelets, and their subtypes. At the same time, it can also accurately detect cell morphological abnormalities, such as uneven size, heteromorphism, nuclear and cytoplasmic changes, etc., and output the cell type corresponding to the cell image. In a specific implementation, it can also output information such as abnormal characteristics corresponding to the cell image.

[0101] In the embodiment of the present application, the cell type corresponding to each cell image is determined to be multiple and each cell type generated corresponds to an accurate probability. For example, a cell image is determined to be neutrophils -75%, eosinophils -25%, and basophils -0%.

[0102] In a specific embodiment, step S100, based on the recognition model, recognizes the blood smear image, and also includes an image preprocessing step. Specifically, it includes performing noise reduction, contrast enhancement, grayscale correction and other operations on the acquired blood smear image, removing the noise interference generated during the image acquisition process, highlighting the morphological details of blood cells, making the image features easier to extract, and then outputting the preprocessed blood smear image to the recognition model. When the acquired blood smear image to be recognized is not clear, the above steps can improve the recognition accuracy of the later recognition model and improve the reliability of the counting results.

[0103] Combination Figure 3 As shown in the figure, for the same type of cells, the blood smear image is divided into multiple different probability areas, and the probability of each type of cell appearing at different positions on the blood smear can be obtained by theoretical analysis or big data statistics. Blood cells of different sizes and shapes will have different probabilities of appearing at different positions on the slide when the slide pushes the blood at different moving speeds or tilt angles.

[0104] In step S300, in order to unify the coordinate system for each blood cell morphology screening and analysis, in actual application, the blood sample for analysis will be dropped at the same position on the glass slide, and the same blood smear making parameters will be used when making the blood smear, such as the speed of the pusher, the tilt angle, etc. Then, the position where the blood sample is dropped, that is, the initial position, is used as the origin of the plane two-dimensional coordinate system, and the length and width of the glass slide are used as the x-axis and y-axis, and the position coordinates corresponding to each cell image are counted.

[0105] Step S300, assigning reliability weights to the accurate probabilities corresponding to each cell type in combination with the probability distribution diagram, and generating reliable values ​​for type determination corresponding to each cell image. In a simple implementation, the accurate probability corresponding to each cell type is directly superimposed with the occurrence probability of the coordinates of its position, to obtain reliable values ​​for type determination corresponding to each cell image. For example, for an A cell image, after image recognition, the probability of being determined as an eosinophil is 70%, and the probability of being determined as a neutrophil is 30%. At the location of the A cell image, the probability of eosinophils appearing is 20%, and the probability of neutrophils appearing is 30%. The type determination reliability value of the A cell image finally determined as an eosinophil is 0.9, and the type determination reliability value of the A cell image determined as a neutrophil is 0.5.

[0106] In step S400, the type determination reliability value obtained in step S300 is compared with a set threshold value, such as 0.8, and finally the cell image is determined to be a sample of a determined type according to the comparison result. In practical applications, the set threshold values ​​corresponding to each type of cell are not the same, and the set threshold values ​​corresponding to each type of cell can be set based on a large amount of determination data and the accuracy of the determined sample.

[0107] If the cell type corresponding to the cell image cannot be determined after S410, it is usually confirmed by manual microscopic examination, and can also be further determined by an extended determination step. Finally, the counting is performed based on the labeling information of each cell image in the blood smear image.

[0108] When screening blood cell morphology, the main reason affecting the accuracy of screening and counting is the overlapping cell images. Figure 1 As shown, in the blood smear image, some cells will overlap partially or even completely cover each other. When the recognition model is used to analyze the above cell overlapping images, there will often be deviations. For this reason, it is necessary to supplement that the cell images described in the embodiments of the present application include single cell images and cell overlapping images. The recognition model stores the association between single cell images and cell types, and between cell overlapping images and the cell types and quantity combinations constituting the above cell overlapping images, which are stored as a first relationship database.

[0109] In step S100, determining and labeling the cell type and accurate probability corresponding to each cell image includes:

[0110] Obtain a single cell image and a cell overlapping image in a blood smear image, search for matching cell images similar to the single cell image and the cell overlapping image based on the first relational database, obtain the corresponding cell type or cell type and quantity combination, and mark the accuracy probability for each matching result according to the similarity between the cell images.

[0111] In the embodiment of the present application, the output of a single cell image is the cell type and its accurate probability, while the cell overlap image corresponds to a combination of at least two cell types, and the number of cells of each type in the above combination may be different. For example, a cell overlap image corresponds to a cell composition including two red blood cells and one white blood cell, that is, the above cell overlap image corresponds to a combination of pairs of cells. Similarly, in order to facilitate the later reliability determination, the accurate probability will also be marked for the above combination.

[0112] In the embodiment of the present application, in order to further improve the reliability of the blood cell counting result, the blood cell counting method further includes:

[0113] The correlation between the overlapping probability of each type of cells, the cell overlapping position and the blood smear production parameters is established and stored as a second relational database. The above correlation can be obtained by training and learning big data through a deep neural network.

[0114] In step S420, the extension determination step includes:

[0115] S421, obtain a cell overlapping image, obtain multiple corresponding cell type and quantity combinations through the first relational database, and configure a first accurate probability value P1 for each combination. For example, for a cell overlapping image, the determination result according to the first relational database may be: a combination of 1 eosinophil and 1 red blood cell, the corresponding first accurate probability value P1 is 80%, or a combination of 1 eosinophil and 2 red blood cells, the corresponding first accurate probability value P1 is 20%.

[0116] S422, obtaining the position coordinates of the cell overlapping image, configuring the second accurate probability value P2 for each of the cell type and quantity combinations according to the second relational database and the probability distribution diagram, and configuring the third accurate probability value P3 for the cell type in the combination. The second accurate probability value P2 is the probability value of the combination of cells corresponding to the cell overlapping image, and the third accurate probability value P3 is the probability value of a single type of cell in the combination. For example, according to the position coordinates of the cell overlapping image, the second accurate probability value P2 of the combination of 1 neutrophil and 1 platelet is 60%, while the probability of a single neutrophil appearing at the above position coordinates is 30%, and the probability of a platelet appearing at the above position coordinates is 90%.

[0117] S423, obtaining blood smear making parameter data corresponding to the current blood smear image, and configuring a fourth accurate probability value P4 for each of the cell type and quantity combinations according to the second relational database. The blood smear making parameters include the angle between the pusher and the slide, the moving speed of the pusher, the ambient temperature in the instrument, and the blood volume.

[0118] S424, assign reliable weight values ​​to the first accurate probability value P1, the second accurate probability value P2, the third accurate probability value P3, and the fourth accurate probability value P4, and calculate and obtain the cell type, quantity, and class quantity determination reliability value K corresponding to each cell overlapping image according to the set algorithm. In one embodiment, the above-mentioned class quantity determination reliability value K=0.8P1+0.3P2+0.4P3+0.2P4, and the class quantity determination reliability value corresponding to each type of cell or its combination can be obtained by calculation through the above formula, and then the cell type and quantity with the largest class quantity determination reliability value are taken as the cell type and quantity corresponding to the cell overlapping image.

[0119] S425, comparing the above-mentioned reliable value of the class quantity determination with a set threshold value:

[0120] S4251, if the reliable value of the class quantity determination corresponding to the cell overlapping image exceeds the set threshold, it is determined to be a confirmed sample and its corresponding cell type and quantity are marked;

[0121] S4252, if the reliability value of the class quantity determination corresponding to the cell overlapping image does not exceed the set threshold, it is determined as a sample to be determined, and the cell type and quantity of the sample to be determined are manually confirmed and marked.

[0122] The above technical solution introduces the factors that affect the probability of cell overlap and the position of cell overlap into the determination of the cell type of the sample to be determined, which can effectively improve the accuracy of the determination of the cell type corresponding to the sample to be determined, and provide reliable data support for the doctor's subsequent diagnosis.

[0123] In practical applications, the distribution of various types of blood cells on a blood smear usually changes evenly and continuously, that is, the ratio of the number of each type of cells in each statistical area should be a smooth curve. Based on the above distribution characteristics, in the implementation mode of the present application, Figure 4 As shown, the blood cell counting method also includes:

[0124] S430, dividing the whole or part of the blood smear image into a plurality of continuous statistical regions of equal area, wherein the statistical regions are preferably rectangular and arranged along the length direction of the blood smear, wherein the criterion for dividing the statistical regions is that the sum of the differences in the number of samples to be determined in each adjacent statistical region is the largest.

[0125] S431, counting the number of each type of cells contained in the determined samples in each statistical area, and calculating the proportion of the number of each type of cells in the statistical area, for example, red blood cells account for 60%, white blood cells account for 35%, platelets account for 5%, among which neutrophils account for 50%, lymphocytes account for 20%, monocytes account for 3%, etc.

[0126] S432, according to the ratio of the number of set type of cells in each statistical area, a trend curve of the ratio changing with the statistical area is generated by fitting, wherein the horizontal axis of the trend curve is the number information of the statistical area, and the vertical axis is the ratio of the number of set type of cells.

[0127] S433, obtaining the cell type and quantity combination corresponding to the cell overlapping image in the statistical area, and decomposing the cell type and quantity combination into the estimated quantity of each type of cells, for example, decomposing a cell overlapping image into one red blood cell and one neutrophil.

[0128] S434, adding the estimated number of each type of cells to the statistical data of the current statistical area, and calculating the ratio of the number of each type of cells in the current statistical area before and after the estimated number is added.

[0129] S435, configuring a fifth accurate probability value for each of the cell type and quantity combinations according to the degree of deviation between the quantity proportion ratio and the trend curve. In practical applications, the fifth accurate probability value is inversely proportional to the degree of deviation, that is, the greater the degree of deviation caused by adding the disassembled cell type and quantity to the statistical area, the lower the reliability of the disassembled cell type and quantity.

[0130] S436, assign weight values ​​to the first accurate probability value, the second accurate probability value, the third accurate probability value, the fourth accurate probability value, and the fifth accurate probability value, calculate and obtain the cell type, quantity, and reliable value of the class quantity determination contained in the cell overlapping image according to the set algorithm, and finally determine the final cell type and quantity according to the reliable value of the class quantity determination. In the embodiment of the present application, the calculation method of step S436 is the same as that of step S424, and the weight value can be fine-tuned according to the actual situation to make the final determination result more accurate, and the specific process will not be repeated here.

[0131] Through the above technical solution, after the number of cells of each type inferred from the cell overlapping image is added to the existing number of cells of each type, if the ratio of the number of cells of each type changes too much, that is, the ratio of the number in the current statistical area deviates greatly from the trend curve, it can be inferred that the number of cells of each type inferred by reverse inference does not meet the statistical probability distribution law, so the reliability of the inference is greatly reduced, and vice versa. At the same time, even if some cell images in the blood smear image are difficult to determine their original cell composition due to cell overlap, it is possible to assign corresponding reliability weight values ​​to various cell types and number combinations through probability distribution, thereby further improving the reliability of cell type and number statistics.

[0132] In order to facilitate the inspector to quickly review the number and morphology of blood cells of a set type, the cell number counting method described in the present application also includes the following steps: receiving and responding to an external image review request, querying the target cell type and enhancing or fading the set cell image based on the annotation information of each cell image in the blood smear image. If the inspector inputs the cell type that needs to be confirmed again, the system will then brighten and magnify the above-mentioned type of cells on the display screen, so that the inspector can review the above-mentioned cells more quickly, thereby improving the efficiency of the entire blood cell counting.

[0133] In order to implement the above-mentioned blood cell counting method based on image recognition, the embodiment of the present application also discloses a blood cell counting system 400 based on image recognition, such as Figure 5 As shown, it mainly includes the following functional units: a data storage unit 1, an image recognition unit 2, a reliability evaluation unit 3 and a count generation unit 4.

[0134] The data storage unit 1 is configured to acquire and store the blood smear image data to be identified, and a probability distribution diagram for characterizing the position distribution law of different types of blood cells on the blood smear. In practical applications, the data storage unit 1 is configured as a built-in data storage hard disk of a PC or an external database. The data storage unit 1 is connected to an external image acquisition device, such as a high-precision microscope imaging device, through a data interface, and can quickly and clearly acquire blood cell images of the blood smear. During the acquisition process, uniform illumination and precise focal length are ensured to obtain high-quality original image data for subsequent analysis. In one embodiment, the blood cell number counting system 400 described in the present application may include the above-mentioned image acquisition device. In another embodiment, the cell number counting system described in the present application may also be configured in a cloud server to directly receive the blood smear images uploaded by each terminal for analysis, and then send the analysis results to each terminal.

[0135] The image recognition unit 2 has a built-in recognition model, and is configured to be data-connected with the data storage unit 1, retrieve the blood smear image data, and recognize the blood smear image based on the recognition model, and determine and mark the cell type, accurate probability and position coordinates corresponding to each cell image. In practical applications, the image recognition unit 2 is also configured with an image preprocessing module, which is used to perform operations such as noise reduction, contrast enhancement, and grayscale correction on the collected original image, remove the noise interference generated during the image acquisition process, highlight the morphological details of blood cells, and make image features easier to extract. The preprocessed image enters the recognition model for recognition. In a specific embodiment, the recognition model is configured in the cloud server, and the above-mentioned image preprocessing process can be performed on each terminal, and then the preprocessed blood smear image is uploaded to the recognition model in the cloud server for recognition processing.

[0136] The reliability evaluation unit 3 is configured to be data-connected with the image recognition unit 2, and is used to obtain the cell type and accurate probability corresponding to each cell image after judgment, as well as the position coordinates of each cell image, and then assign reliability weights to the accurate probabilities in combination with the probability distribution diagram, and generate type judgment reliability values ​​corresponding to each cell image.

[0137] The counting generation unit 4 has a built-in comparison output module, which is configured to be data-connected with the reliability evaluation unit 3, receive the type determination reliability value corresponding to each cell image and compare it with a set threshold value: if it exceeds the set threshold value, the corresponding cell image is determined as a confirmed sample and its cell type is marked; if it does not exceed the set threshold value, it is determined as a sample to be determined and the cell type and quantity of the above sample to be determined are confirmed and marked by manual and / or extended determination steps. Finally, based on the annotation information of each cell image in the blood smear image, the number of each type of blood cells is counted and output. In actual applications, the above set threshold value can be fine-tuned by the inspector according to the actual situation to make the final determination result more accurate.

[0138] In order to further improve the reliability of the recognition results, the data storage unit 1 also stores: a first association relationship between the overlap probability of each type of cells, the cell overlap position and the blood smear production parameters, and a second association relationship between the cell overlap image and the cell type and quantity combination constituting the above cell overlap image.

[0139] The blood cell counting system 400 described in the present application further includes an extended data preprocessing unit 5 and a first extended determination unit 6 .

[0140] The extended data preprocessing unit 5 is configured to be connected with the data storage unit 1, the image recognition unit 2 and the external smear instrument data for automatically making blood smears, obtain the cell overlapping image and obtain multiple cell types and quantity combinations corresponding to the above-mentioned cell overlapping image according to the above-mentioned first association relationship, and configure the first accurate probability value for each combination. At the same time, the position coordinates of the cell overlapping image are obtained, and the second accurate probability value is configured for each of the cell types and quantity combinations according to the second association relationship, and then the third accurate probability value is configured for each cell type in the combination according to the above-mentioned probability distribution diagram. Finally, the blood smear production parameter data corresponding to the current blood smear image is obtained from the smear instrument, such as the pushing speed and angle, and the fourth accurate probability value is configured for each cell type and quantity combination according to the second association relationship. In a specific embodiment, the production of the blood smear is completed manually. For this reason, the system is also configured with a blood smear production parameter input unit, and the corresponding blood smear production parameters are manually input.

[0141] The first extension determination unit 6 is configured to be data-connected with the extension data preprocessing unit 5, and assigns reliable weight values ​​to the first accurate probability value, the second accurate probability value, the third accurate probability value, and the fourth accurate probability value. Then, the cell type, quantity, and reliable value for class quantity determination corresponding to the cell overlapping image are calculated according to the set algorithm. Finally, the reliable value for class quantity determination is compared with a set threshold value, and the cell type and quantity corresponding to the cell overlapping image are determined according to the comparison result, or the cell type and quantity corresponding to the cell overlapping image are determined manually.

[0142] In the specific implementation process, the extended data pre-processing unit 5 and the first extended determination unit 6 both realize specific functions through specific computer program modules.

[0143] In the above technical solution, the blood smear production parameters that affect the cell distribution and overlap probability on the blood smear are incorporated into the determination probability analysis of uncertain samples, which can significantly improve the reliability of the determination results.

[0144] Further optimized, the blood cell number counting system 400 also includes an image area segmentation unit 7, a number ratio statistics unit 8, a reference curve generation unit 9, a disassembly and comparison unit 10 and a second extension determination unit 11.

[0145] The image region segmentation unit 7 is configured to divide the whole or part of the blood smear image into a plurality of continuous statistical regions of equal area, and the statistical regions are preferably rectangular and arranged along the length direction of the blood smear. The number proportion statistical unit 8 is configured to be connected with the counting generation unit 4 for counting the number of each type of cells contained in the determined sample in each statistical region, and calculating the number proportion of each type of cells in the statistical region. The reference curve generation unit 9 is configured to fit and generate a trend curve of the above ratio changing with the statistical region according to the number proportion ratio of the set type of cells in each statistical region. The disassembly and comparison unit 10 is configured to be connected with the extended data preprocessing unit 5 for obtaining the cell type and number combination corresponding to the cell overlapping image of the statistical region, disassembling the above cell type and number combination into the estimated number of each type of cells, and then adding the estimated number of each type of cells to the statistical data of the current statistical region, calculating the number proportion ratio of each type of cells in the current statistical region before and after the above estimated number is added, and configuring the fifth accurate probability value for each cell type and number combination according to the degree of deviation of the above number proportion ratio from the trend curve.

[0146] The second extended determination unit 11 is used to assign weight values ​​to the first accurate probability value, the second accurate probability value, the third accurate probability value, the fourth accurate probability value and the fifth accurate probability value, and then calculate and obtain the cell type and quantity contained in the cell overlapping image according to the set algorithm. The specific implementation process has been recorded in the method embodiment and will not be repeated here.

[0147] In the embodiment of the present application, a blood cell counting device based on image recognition is also disclosed, such as Figure 6 As shown, the device mainly includes: an image acquisition device 100, a blood smear production parameter acquisition device 200, an information interaction device 300 and a blood cell number counting system 400 based on image recognition as described above.

[0148] In detail, the image acquisition device 100 is configured to acquire and output blood smear images. The above-mentioned image acquisition device 100 can be configured as a high-precision microscope imaging device, or directly configured as an image data acquisition interface, and the blood smear image data to be identified is shared by other terminals.

[0149] The blood smear making parameter acquisition device 200 is configured to be connected to the data of the smear machine that automatically makes blood smears, and is used to obtain and output the blood smear making parameter data. In a specific embodiment, the blood smear making parameter acquisition device 200 also includes a parameter manual input module configured in the information interaction device 300, and the inspection personnel manually input the blood smear making parameters.

[0150] The blood cell number counting system 400 is configured to be data-connected with the image acquisition device 100 and the blood smear production parameter acquisition device 200, and is used to obtain blood smear images and blood smear production parameter data, and output the number of each type of blood cells and annotation information after processing by the built-in program algorithm module. In actual applications, the main body of the above-mentioned blood cell number counting system 400 can be configured in a cloud server, and directly configured as a data interaction module in a local blood cell number counting device, which receives the blood smear image data uploaded by the user after verifying the user's authority and then feeds back the corresponding counting and analysis results according to the needs.

[0151] The information interaction device 300 is configured to be data-connected to the blood cell counting system 400, and is used to receive the user's operating instructions and display and output the various types of blood cell counts and annotation information obtained from the system end. In actual applications, the above-mentioned information interaction device 300 can be configured as a PC or a tablet computer with network data communication function.

[0152] In order to facilitate the promotion and use of the blood cell counting method involved in the embodiments of the present application, a computer-readable storage medium is also proposed in the present application, on which a program module for implementing the blood cell counting method based on image recognition as described above is loaded. In practical applications, the above-mentioned computer-readable storage medium includes but is not limited to a disk storage, a CD-ROM, an optical storage, etc.

[0153] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A blood cell counting method based on image recognition, characterized in that: include: Identify blood smear images based on the recognition model, determine and label the cell type and accurate probability corresponding to each cell image; Obtaining and generating a probability distribution diagram for characterizing the probability of each type of blood cell appearing at different positions on the blood smear based on the position distribution rules of different types of blood cells on the blood smear; Obtaining the cell type and accurate probability corresponding to each cell image after determination, as well as the position coordinates of each cell image, assigning a reliability weight to the accurate probability in combination with the probability distribution graph, and generating a type determination reliability value corresponding to each cell image; Compare the type determination reliability value corresponding to each cell image with a set threshold: If the type determination reliability value corresponding to the cell image exceeds the set threshold, the sample is determined to be confirmed and its cell type is marked; If the type determination reliability value corresponding to the cell image does not exceed the set threshold, it is determined as a sample to be determined, and the cell type and quantity of the sample to be determined are confirmed and marked by manual and / or extended determination steps; Based on the annotation information of each cell image in the blood smear image, the number of each type of blood cells is counted and output.

2. The method for counting blood cell number based on image recognition according to claim 1, characterized in that: The cell images include single cell images and cell overlapping images; The recognition model stores the association relationship between a single cell image and a cell type, and between a cell overlapping image and a combination of cell types and quantities constituting the cell overlapping image, which is stored as a first relationship database; In the blood cell counting method, the cell type corresponding to each cell image is determined to be multiple and each generated cell type is determined to correspond to an accurate probability; Determine and label the cell type and accurate probability corresponding to each cell image, including: Obtain a single cell image and a cell overlapping image in a blood smear image, search for matching cell images similar to the single cell image and the cell overlapping image based on the first relational database, obtain the corresponding cell type or cell type and quantity combination, and mark the accuracy probability for each matching result according to the similarity between the cell images.

3. The method for counting blood cell number based on image recognition according to claim 2, characterized in that: The blood cell counting method further comprises: Establishing the correlation between the overlapping probability of each type of cells, the overlapping position of cells and the blood smear making parameters, and storing them as a second relational database; The extension determination step comprises: Acquire overlapping cell images, acquire multiple corresponding cell type and quantity combinations through a first relational database, and configure a first accurate probability value for each combination; Acquire the position coordinates of the cell overlapping images, configure a second accurate probability value for each of the cell type and quantity combinations according to the second relational database and the probability distribution map, and configure a third accurate probability value for the cell type in the combination; Acquire blood smear making parameter data corresponding to the current blood smear image, and configure a fourth accurate probability value for each of the cell type and quantity combinations according to the second relational database; Assigning reliable weight values ​​to the first accurate probability value, the second accurate probability value, the third accurate probability value and the fourth accurate probability value, and obtaining reliable values ​​for determining the cell type, quantity and class quantity corresponding to each cell overlapping image according to a set algorithm; Compare the above-mentioned reliable value of class quantity determination with a set threshold: If the reliable value of the class quantity judgment corresponding to the cell overlapping image exceeds the set threshold, it is determined to be a confirmed sample and its corresponding cell type and quantity are marked; If the reliable value of the class quantity judgment corresponding to the cell overlapping image does not exceed the set threshold, it is judged as a sample to be determined, and the cell type and quantity of the sample to be determined are manually confirmed and marked; The blood smear production parameters include the angle between the pusher and the slide, the moving speed of the pusher and the blood volume.

4. The method for counting blood cell number based on image recognition according to claim 3, characterized in that: The blood cell counting method further comprises: Dividing the whole or part of the blood smear image into a plurality of continuous statistical regions with equal areas; Count the number of each type of cells contained in the determined sample in each statistical area, and calculate the proportion of each type of cells in the statistical area; According to the ratio of the number of set type of cells in each statistical area, a trend curve of the ratio changing with the statistical area is generated by fitting; Obtain cell types and quantity combinations corresponding to overlapping images of cells in the statistical area, and decompose the above cell types and quantity combinations into estimated quantities of cells of each type; The estimated number of cells of each type is added to the statistical data of the current statistical area, and the ratio of the number of cells of each type in the current statistical area before and after the estimated number is added is calculated; According to the degree of deviation between the above-mentioned quantity proportion ratio and the above-mentioned trend curve, a fifth accurate probability value is configured for each of the above-mentioned cell type and quantity combinations; Assigning weight values ​​to the first accurate probability value, the second accurate probability value, the third accurate probability value, the fourth accurate probability value and the fifth accurate probability value, and calculating and obtaining the cell types and quantities contained in the cell overlapping image according to a set algorithm; The criterion for dividing the statistical area is that the sum of the differences in the number of samples to be determined in each adjacent statistical area is the largest.

5. The method for counting blood cell number based on image recognition according to claim 1, characterized in that: The cell number counting method further comprises: Receive and respond to external image review requests, query the target cell type based on the annotation information of each cell image in the blood smear image, and enhance or fade the set cell image.

6. A blood cell counting system based on image recognition, characterized in that: include: A data storage unit (1) configured to acquire and store blood smear image data to be identified and a probability distribution diagram for characterizing the position distribution law of different types of blood cells on the blood smear; An image recognition unit (2) has a built-in recognition model and is configured to be data-connected to the data storage unit (1), retrieve the blood smear image data and recognize the blood smear image based on the recognition model, and determine and mark the cell type, accurate probability and position coordinates corresponding to each cell image therein; A reliability evaluation unit (3) is configured to be data-connected to the image recognition unit (2) and is used to obtain the cell type and accuracy probability corresponding to each cell image after determination, as well as the position coordinates of each cell image, assign a reliability weight to the accuracy probability in combination with the probability distribution diagram, and generate a type determination reliability value corresponding to each cell image; The counting generation unit (4) has a built-in comparison output module and is configured to be data-connected to the reliability evaluation unit (3), receive the type determination reliability value corresponding to each cell image and compare it with a set threshold value: if it exceeds the set threshold value, the corresponding cell image is determined as a confirmed sample and its cell type is marked; If it does not exceed the set threshold, it is determined as a sample to be determined, and the cell type and quantity of the sample to be determined are confirmed and marked by manual and / or extended determination steps; based on the marking information of each cell image in the blood smear image, the number of each type of blood cells is counted and output.

7. The blood cell counting system based on image recognition according to claim 6, characterized in that: The data storage unit (1) also stores: a first correlation between the probability of cell overlap of each type, the cell overlap position and the blood smear production parameters, and a second correlation between the cell overlap image and the cell type and quantity combination constituting the cell overlap image; The blood cell counting system (400) further comprises: The extended data preprocessing unit (5) is configured to be data-connected with the data storage unit (1) and the image recognition unit (2), obtain the cell overlapping image and obtain a plurality of cell type and quantity combinations corresponding to the cell overlapping image according to the first association relationship, and configure a first accurate probability value for each combination; obtain the position coordinates of the cell overlapping image, configure a second accurate probability value for each cell type and quantity combination according to the second association relationship, and configure a third accurate probability value for each cell type in the combination according to the probability distribution diagram; obtain the blood smear production parameter data corresponding to the current blood smear image, and configure a fourth accurate probability value for each cell type and quantity combination according to the second association relationship; The first extended determination unit (6) is configured to be data-connected to the extended data preprocessing unit (5), assign reliable weight values ​​to the first accurate probability value, the second accurate probability value, the third accurate probability value and the fourth accurate probability value, and calculate and obtain the cell type, quantity and class quantity determination reliability value corresponding to the cell overlapping image according to a set algorithm; compare the class quantity determination reliability value with a set threshold value, and determine the cell type and quantity corresponding to the cell overlapping image according to the comparison result, or determine the cell type and quantity corresponding to the cell overlapping image by manual determination.

8. The blood cell counting system based on image recognition according to claim 7, characterized in that: The blood cell number counting system also includes: An image region segmentation unit (7) configured to divide the entire or partial region of the blood smear image into a plurality of continuous statistical regions of equal area; A number proportion statistical unit (8) is configured to be data-connected to the count generation unit (4) and is used to count the number of each type of cells contained in the determined sample in each statistical area and calculate the number proportion of each type of cells in the statistical area; A reference curve generating unit (9) is configured to generate a trend curve of the ratio changing with the statistical area according to the ratio of the number of set type of cells in each statistical area; The disassembly and comparison unit (10) is configured to be data-connected to the extended data preprocessing unit (5), and is used to obtain the cell type and quantity combination corresponding to the cell overlapping image of the statistical area, disassemble the above cell type and quantity combination into the estimated quantity of each type of cells, add the estimated quantity of each type of cells to the statistical data of the current statistical area, calculate the proportion of the number of each type of cells in the current statistical area before and after the above estimated quantity is added, and configure a fifth accurate probability value for each of the cell type and quantity combinations according to the degree of deviation between the above quantity proportion ratio and the trend curve; The second extension determination unit (11) is used to assign weight values ​​to the first accurate probability value, the second accurate probability value, the third accurate probability value, the fourth accurate probability value and the fifth accurate probability value, and calculate and obtain the cell types and quantities contained in the cell overlapping image according to a set algorithm.

9. A blood cell counting device based on image recognition, characterized in that: include: An image acquisition device (100) configured to acquire and output a blood smear image; A blood smear making parameter acquisition device (200), configured to be data-connected to a smear machine for automatically making a blood smear, for acquiring blood smear making parameter data and outputting it; The blood cell counting system (400) based on image recognition as claimed in any one of claims 6 to 8 is configured to be data-connected to the image acquisition device (100) and the blood smear production parameter acquisition device (200), and is used to acquire blood smear images and blood smear production parameter data, and output the number of each type of blood cells and labeling information after processing by a built-in program algorithm module; as well as The information interaction device (300) is configured to be data-connected to the blood cell counting system (400) and is used to receive user operation instructions and display and output the blood cell counts and labeling information of various types obtained from the system end.

10. A computer-readable storage medium, characterized in that: A program module for implementing the blood cell counting method based on image recognition as described in any one of claims 1 to 5 is loaded thereon.

Images