Blood cell analysis method and system based on microscopic imaging

Abstract

The invention discloses a blood cell analysis method and system based on microscopic imaging, and particularly relates to the technical field of blood cell analys.The blood cell overlapping degree CO of a blood cell imaging picture is obtained, and a smear uniformity coefficient TO is obtained through calculation based on position coordinates of a connected region; jointly analyzing the smear uniformity coefficient and the blood cell overlapping degree to obtain a smear quality evaluation index TQ; the relation between the smear quality evaluation index and the threshold value is judged, and the problem of human resource waste is solved; identifying an error to-be-analyzed region in the blood cell imaging picture to obtain a region high-definition image corresponding to the error to-be-analyzed region; respectively inputting the blood cell imaging picture and the regional high-definition image into a convolutional neural network, outputting blood cell feature information, and summarizing the blood cell feature information to obtain corrected blood cell feature information; inaccurate analysis results caused by errors can be avoided; and the problems that the data size is too large and too many analysis resources are occupied due to high-definition analysis on the smear are also avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of blood cell analysis, and more particularly, to a blood cell analysis method and system based on microscopic imaging. Background Art

[0002] Microscopic blood cell analysis involves imaging blood cells using a microscope and using image processing and analysis techniques to identify, classify, and count them. This method has widespread applications in medical diagnosis, biomedical research, and other fields.

[0003] The blood cell analysis method based on microscopic imaging includes: a high-resolution optical microscope for obtaining clear blood cell images, the optical microscope having adjustable magnification and focal length, and a stable light source; a control system for automatic movement and focusing of the microscope: a control system for automatic movement and focusing of the microscope that can automatically move the microscope stage and adjust the focal length; an image acquisition device for converting cell images under the microscope into digital signals and transmitting them to a computer for processing;

[0004] However, in actual use, it still has many shortcomings. For example, abnormal smear quality leads to uneven distribution of blood cells in the output blood cell imaging image, a large number of blood cells overlap, and inaccurate analysis is impossible. In the existing technology, reliable blood cell imaging images are often obtained through manual screening, which increases the consumption of human resources and is not conducive to the automated analysis of blood cells. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a blood cell analysis method and system based on microscopic imaging to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a blood cell analysis method based on microscopic imaging, comprising the following steps:

[0007] Step 1: Obtain several microscopic images of the smear by rotating the coarse focusing screw and the fine focusing screw, and output a blood cell imaging image;

[0008] Step 2: Obtain the blood cell overlap CO of the blood cell imaging image, and calculate the smear uniformity coefficient TO based on the position coordinates of the connected area; jointly analyze the smear uniformity coefficient and the blood cell overlap to obtain the smear quality assessment index TQ; calculate the smear quality assessment index TQ using the formula TQ = w1*TO + W2*(1-CO), where w1 represents the influence coefficient of the smear uniformity coefficient and w2 represents the influence coefficient of the overlap;

[0009] Step 3: Determine the relationship between the smear quality assessment index and the threshold;

[0010] Step 4: Blood cell imaging image feature analysis: Input the blood cell imaging image into the convolutional neural network, output the blood cell feature information, and mark the blood cell imaging image;

[0011] Step 5. Disease risk auxiliary analysis: The correlation between the disease and blood cell characteristic information is obtained through correlation analysis. The disease risk probability is matched for each smear based on the correlation, and a disease auxiliary diagnosis model is built.

[0012] Preferably, the blood cell overlap degree of the blood cell imaging picture is obtained by:

[0013] Step S01: converting the blood cell imaging picture into a grayscale image, and using Gaussian filtering and smoothing technology to remove image noise; converting the blood cell imaging picture into a binary image;

[0014] Step S02, morphological analysis: using morphological operations and connected region labeling algorithms to identify each blood cell or blood cell group in the image and mark connected regions; there are m connected regions, and j represents the number of connected regions;

[0015] Step S03: Get the area of ​​each connected region, record the area of ​​the j-th connected region as ASj, and use the formula The total blood cell area Ato was calculated;

[0016] Step S04, identifying overlapping regions and calculating the area of ​​overlapping regions: setting the distance between the centers of blood cells to be less than the overlap judgment threshold, the blood cells are considered to be overlapping and the overlapping regions are marked; calculating the area of ​​overlapping regions through morphological operations, recording the overlapping area of ​​the j-th connected region as ACj, and using the formula The overlapping area Atv is calculated;

[0017] Step S05: By formula The blood cell overlap degree CO of the blood cell imaging image is calculated.

[0018] Preferably, the smear uniformity coefficient is obtained by: obtaining the two-dimensional position coordinates of each connected area on the smear through image recognition technology; dividing the smear into multiple grids of equal area, and calculating the number of connected areas in each grid; calculating the density of the connected areas in each grid, and expressing the smear uniformity coefficient TO as the inverse of the variance of the connected area density.

[0019] Preferably, when the smear quality assessment index is lower than a threshold, an early warning is issued to prompt the smear to be replaced; when the smear quality assessment index is not lower than the threshold, a blood cell imaging picture qualified instruction is output; based on the blood cell imaging picture qualified instruction, an imaging picture feature analysis instruction is started.

[0020] Preferably, the convolutional neural network is a trained blood cell type recognition model, and the process of building the blood cell type recognition model includes the following steps:

[0021] Step S11: obtaining a blood cell imaging image of a smear, manually marking the actual characteristic information of each blood cell, and obtaining a sample data set;

[0022] Step S12: using the blood cell imaging image as input to a blood cell type recognition model, and outputting predicted feature information of each blood cell;

[0023] Step S13, model initialization: defining the initial parameters of deep learning, including weight parameters, bias parameters, and activation functions between neural networks;

[0024] Step S14: Divide the sample data set into model training samples and test samples according to the proportion, input the training samples into the convolutional neural network, update the inter-neural network weights, bias parameters and activation functions of the neural network through forward propagation and backward propagation of the neural network, calculate the loss function, and set the loss function based on the similarity between the predicted feature information and the actual feature information;

[0025] Step S15: If the loss value does not meet the preset threshold requirement, the weight parameters and bias parameters need to be updated through the back propagation algorithm; the back propagation algorithm calculates the gradient of the loss function with respect to each weight and bias, that is, the partial derivative of the loss value with respect to these parameters, and updates the value of each parameter based on the gradient information;

[0026] Step S16, iterative training: Repeat the steps of forward propagation, loss calculation, backpropagation, and parameter update until the loss value meets the preset threshold requirement or reaches a predetermined number of iterations, thereby obtaining a trained blood cell type recognition model; verify the accuracy and precision of the model through test samples, and deploy the trained blood cell type recognition model into the application.

[0027] Preferably, the construction of the disease-assisted diagnosis model includes the following steps:

[0028] Step S21, data collection and preprocessing: obtaining blood cell characteristic information corresponding to each disease type, checking and processing missing values, abnormal values ​​and duplicate records, and outputting a sample set;

[0029] Step S22, data partitioning: Divide the sample set into a training set, a validation set, and a test set; the training set is used to train the decision tree, allowing the decision tree to learn the mapping relationship from blood cell feature information to the target variable; the validation set is used to adjust the parameters of the decision tree, including the number of trees in the random forest, the maximum depth of the tree, and the maximum number of features considered when splitting a node; the test set is used to evaluate the performance indicators of the decision tree;

[0030] Step S23, constructing a random forest: a decision tree algorithm is selected to construct a single decision tree; when constructing each tree, a subset is extracted from the training set through sampling with replacement as the training data for the tree; when splitting each node of the tree, only the randomly selected feature subset is considered; the above process is repeated to construct multiple decision trees to form a random forest; the prediction results of multiple trees are integrated using majority voting or regression method;

[0031] Step S24: Use the test set to evaluate the performance of the random forest. The obtained evaluation indicators include accuracy, recall rate, F1 score, and area under the ROC curve; adjust the parameters of the random forest through the cross-validation method until the evaluation indicator requirements are met, and output the disease auxiliary diagnosis model.

[0032] Preferably, the blood cell analysis method further comprises:

[0033] Based on the qualified instruction of the blood cell imaging picture, an error risk analysis instruction for the blood cell imaging picture is started;

[0034] Error risk analysis is used to identify error areas in blood cell imaging images. These areas are areas where the risk of blood cell analysis errors is high due to blood cell overlap. By rotating the converter to replace the high-power objective lens and repeatedly adjusting the focus until a clear image is observed under the high-power lens, a high-definition image of the area corresponding to the error area is obtained.

[0035] The blood cell imaging image and the regional high-definition image are respectively input into the convolutional neural network. After the blood cell characteristic information is output, it is summarized and marked on the blood cell imaging image to obtain the corrected blood cell characteristic information. The corrected blood cell characteristic information is input into the disease auxiliary diagnosis model, and the type of disease present in the smear and the corresponding probability are output.

[0036] Preferably, the error area to be analyzed is obtained by dividing the blood cell imaging image into several regions of equal area, with k regions being provided, and s being used to represent the region number; obtaining the overlap degree and the total area and clarity of the connected regions of each region, which are recorded as co_s, as_s, and xs_s respectively; and using the formula The error risk coefficient wf of each area is calculated, where α represents the influence coefficient of the product of clarity and area, β represents the influence coefficient of overlap, and Fnorm(·) represents the linear normalization function, which is used to limit the value of overlap to the range of 0 to 1. Based on the preset error threshold, the error area to be analyzed is screened.

[0037] To achieve the above objectives, the present invention provides the following technical solution: a blood cell analysis system based on microscopic imaging, comprising:

[0038] The blood cell imaging picture acquisition module obtains several microscopic images of the smear by rotating the coarse focusing screw and the fine focusing screw of the microscope, and obtains the blood cell imaging picture corresponding to each smear;

[0039] The smear quality analysis module is used to obtain the smear quality evaluation index TQ of the blood cell imaging image; obtain the blood cell overlap CO of the blood cell imaging image, and calculate the smear uniformity coefficient TO based on the position coordinates of the connected area; and jointly analyze the smear uniformity coefficient and the blood cell overlap to obtain the smear quality evaluation index TQ;

[0040] The smear quality judgment module is used to judge the relationship between the smear quality assessment index and the threshold; when the smear quality assessment index is not lower than the threshold, the blood cell imaging picture qualified instruction is output; based on the blood cell imaging picture qualified instruction, the imaging picture feature analysis instruction is started;

[0041] The blood cell imaging picture feature analysis module is used to obtain the blood cell feature information corresponding to the smear, input the blood cell imaging picture into the convolutional neural network, and output the blood cell feature information;

[0042] The disease risk auxiliary analysis module is used to obtain the correlation between disease and blood cell characteristic information through correlation analysis, match the disease risk probability for each smear based on the correlation, and build a disease auxiliary diagnosis model.

[0043] Preferably, the system further comprises:

[0044] The blood cell characteristic information correction module is used to correct the blood cell characteristic information and use the corrected blood cell characteristic information to assist in disease risk assessment; it includes:

[0045] Based on the qualified instruction of the blood cell imaging picture, an error risk analysis instruction for the blood cell imaging picture is started;

[0046] Error risk analysis is used to identify error areas in blood cell imaging images. These areas are areas where the risk of blood cell analysis errors is high due to blood cell overlap. By rotating the converter to replace the high-power objective lens and repeatedly adjusting the focus until a clear image is observed under the high-power lens, a high-definition image of the area corresponding to the error area is obtained.

[0047] The blood cell imaging image and the regional high-definition image are respectively input into the convolutional neural network. After the blood cell characteristic information is output, it is summarized and marked on the blood cell imaging image to obtain the corrected blood cell characteristic information. The corrected blood cell characteristic information is input into the disease auxiliary diagnosis model, and the type of disease present in the smear and the corresponding probability are output.

[0048] Technical effects and advantages of the present invention:

[0049] (1) The blood cell analysis method based on microscopic imaging provided by the present invention obtains the blood cell overlap CO of the blood cell imaging picture, and calculates the smear uniformity coefficient TO based on the position coordinates of the connected area; jointly analyzes the smear uniformity coefficient and the blood cell overlap to obtain the smear quality assessment index TQ; determines the relationship between the smear quality assessment index and the threshold value, and solves the technical problem of waste of human resources; obtains the correlation between the disease and blood cell characteristic information through correlation analysis, matches the disease risk probability for each smear based on the correlation, and provides doctors with auxiliary diagnosis suggestions based on big data and artificial intelligence, helping doctors to more accurately judge the condition and formulate treatment plans, thereby reducing the waste of medical resources and the economic burden on patients.

[0050] (2) The blood cell analysis method based on microscopic imaging provided by the present invention obtains a high-definition image of the region corresponding to the error-to-be-analyzed region by identifying the error-to-be-analyzed region in the blood cell imaging picture; the blood cell imaging picture and the high-definition image of the region are respectively input into a convolutional neural network, and blood cell characteristic information is output and summarized to obtain corrected blood cell characteristic information; it can avoid inaccurate analysis results caused by errors; and it can also avoid the problem of large data size and excessive analysis resources occupied by high-definition analysis of smears. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 Flow chart of the blood cell analysis method of the present invention.

[0052] Figure 2 This is a structural block diagram of the blood cell analysis system of the present invention. DETAILED DESCRIPTION

[0053] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0054] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0055] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0056] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0057] Example 1

[0058] See Figure 1 The present invention provides a blood cell analysis method flow chart. Figure 1 A blood cell analysis method based on microscopic imaging is shown, comprising the following steps:

[0059] Step 1: Obtain several microscopic images of the smear by rotating the coarse focusing screw and the fine focusing screw, and output a blood cell imaging image;

[0060] Specifically, the prepared blood smear is placed on the stage of the microscope and fixed with a smear clamp; the converter is rotated so that the low-power objective lens is facing the light hole; the reflector and the aperture are adjusted to make the field of view brightest and most uniform; looking at the objective lens from the side, the coarse focusing screw is turned to slowly lower the lens barrel until it is close to the blood smear but not touching it; the coarse focusing screw is used to slowly raise the lens barrel until a blurred image is seen, and the fine focusing screw is turned until the image is clear to obtain a blood cell imaging picture, and the images of the red blood cells, white blood cells, and platelets in the smear are obtained by observing the blood cell imaging picture;

[0061] In a further design, an embodiment of the present invention utilizes an automatic microscope focus control system to move the microscope stage and adjust the focal length to achieve scanning and imaging of the entire smear; the automatic microscope focus control system belongs to the prior art, so the present invention does not make specific limitations on this.

[0062] Step 2: Obtain the blood cell overlap CO of the blood cell imaging image, and calculate the smear uniformity coefficient TO based on the position coordinates of the connected area; jointly analyze the smear uniformity coefficient and blood cell overlap to obtain the smear quality assessment index TQ;

[0063] In a further design, the smear quality evaluation index TQ was calculated by the formula TQ = w1*TO + W2*(1-CO), where w1 represents the influence coefficient of the smear uniformity coefficient and w2 represents the influence coefficient of the overlap coefficient;

[0064] Step 3, smear quality judgment: determine the relationship between the smear quality assessment index and the threshold;

[0065] In a further design, when the smear quality assessment index is lower than a threshold, an early warning is issued to prompt the smear to be replaced; when the smear quality assessment index is not lower than the threshold, a blood cell imaging picture qualified instruction is output; based on the blood cell imaging picture qualified instruction, an imaging picture feature analysis instruction is started;

[0066] Step 4: Blood cell imaging image feature analysis: Input the blood cell imaging image into the convolutional neural network, output the blood cell feature information, and mark the blood cell imaging image;

[0067] It should be further explained in the embodiments of the present invention that the blood cell characteristic information includes at least the specific type, morphological characteristics (such as size, shape, texture, volume, etc.), spectral characteristics (such as color, brightness, etc.), and quantitative characteristics of the blood cells; the quantitative characteristics include red blood cell count, white blood cell count, platelet count, and hemoglobin level;

[0068] Step 5: Disease risk auxiliary analysis: Correlation analysis is used to obtain the correlation between the disease and blood cell characteristics, and the disease risk probability is matched for each smear based on the correlation; including:

[0069] The target variables are clearly defined as the type and probability of disease present in the smear, the blood cell characteristic information is input into the disease-assisted diagnosis model, and the type and corresponding probability of disease present in the smear are output.

[0070] It needs to be further explained in the embodiment of the present invention that the blood cell overlap degree of the blood cell imaging picture is obtained in the following manner:

[0071] Step S01: convert the blood cell imaging image into a grayscale image and use Gaussian filtering and smoothing technology to remove image noise; convert the blood cell imaging image into a binary image, that is, the blood cell area is white and the background is black;

[0072] Step S02, morphological analysis: using morphological operations (such as erosion, dilation, opening, closing) and connected region marking algorithms to identify each blood cell or blood cell group in the image and mark the connected regions; there are m connected regions, and j represents the number of the connected regions;

[0073] Step S03: Get the area of ​​each connected region, record the area of ​​the j-th connected region as ASj, and use the formula The total blood cell area Ato was calculated;

[0074] Step S04, identifying overlapping regions and calculating the area of ​​overlapping regions: setting the distance between the centers of blood cells to be less than the overlap judgment threshold, the blood cells are considered to overlap and the overlapping regions are marked; calculating the area of ​​overlapping regions through morphological operations (such as intersection), recording the overlapping area of ​​the jth connected region as ACj, and calculating the area of ​​the jth connected region as ACj through the formula The overlapping area Atv is calculated;

[0075] Step S05: By formula The blood cell overlap degree CO of the blood cell imaging image is calculated.

[0076] What needs to be further explained in the embodiments of the present invention is that the smear uniformity coefficient is obtained by: obtaining the two-dimensional position coordinates of each connected area on the smear through image recognition technology; dividing the smear into multiple grids of equal area, and calculating the number of connected areas in each grid; calculating the density of the connected areas in each grid, and expressing the smear uniformity coefficient TO as the inverse of the variance of the connected area density.

[0077] It should be further explained in the embodiments of the present invention that the convolutional neural network is a trained blood cell type recognition model. The present invention does not limit the specific type of blood cell type recognition model. For ease of understanding, the embodiments of the present invention provide a process for building a blood cell type recognition model, including the following steps:

[0078] Step S11: obtaining a blood cell imaging image of a smear, manually marking the actual characteristic information of each blood cell, and obtaining a sample data set;

[0079] Step S12: using the blood cell imaging image as input to a blood cell type recognition model, and outputting predicted feature information of each blood cell;

[0080] Step S13, model initialization: defining the initial parameters of deep learning, including weight parameters, bias parameters, and activation functions between neural networks;

[0081] Step S14: Divide the sample data set into model training samples and test samples according to the proportion, input the training samples into the convolutional neural network, update the inter-neural network weights, bias parameters and activation functions of the neural network through forward propagation and backward propagation of the neural network, calculate the loss function, and set the loss function based on the similarity between the predicted feature information and the actual feature information;

[0082] Step S15: If the loss value does not meet the preset threshold requirement, the weight parameters and bias parameters need to be updated through the back propagation algorithm; the back propagation algorithm calculates the gradient of the loss function with respect to each weight and bias, that is, the partial derivative of the loss value with respect to these parameters, and updates the value of each parameter based on the gradient information;

[0083] Step S16, iterative training: Repeat the steps of forward propagation, loss calculation, backpropagation, and parameter update until the loss value meets the preset threshold requirement or reaches a predetermined number of iterations, thereby obtaining a trained blood cell type recognition model; verify the accuracy and precision of the model through test samples, and deploy the trained blood cell type recognition model into the application.

[0084] It should be further explained in the embodiments of the present invention that the construction of the disease-assisted diagnosis model includes the following steps:

[0085] The construction of the disease-assisted diagnosis model includes the following steps:

[0086] Step S21, data collection and preprocessing: obtaining blood cell characteristic information corresponding to each disease type, checking and processing missing values, abnormal values ​​(such as values ​​outside the biological range) and duplicate records, and outputting a sample set;

[0087] Step S22, data partitioning: Divide the sample set into a training set, a validation set, and a test set; the training set is used to train the decision tree, allowing the decision tree to learn the mapping relationship from blood cell feature information to the target variable; the validation set is used to adjust the parameters of the decision tree, including the number of trees in the random forest, the maximum depth of the tree, and the maximum number of features considered when splitting a node; the test set is used to evaluate the performance indicators of the decision tree;

[0088] Step S23, constructing a random forest: a decision tree algorithm is selected to construct a single decision tree; when constructing each tree, a subset is extracted from the training set through sampling with replacement as the training data for the tree; when splitting each node of the tree, only the randomly selected feature subset is considered; the above process is repeated to construct multiple decision trees to form a random forest; the prediction results of multiple trees are integrated using majority voting or regression method;

[0089] Step S24: Use the test set to evaluate the performance of the random forest. The obtained evaluation indicators include accuracy, recall rate, F1 score, and area under the ROC curve; adjust the parameters of the random forest through the cross-validation method until the evaluation indicator requirements are met, and output the disease auxiliary diagnosis model.

[0090] Example 2

[0091] This embodiment provides a blood cell analysis method based on microscopic imaging, further comprising:

[0092] Based on the qualified instruction of the blood cell imaging picture, an error risk analysis instruction for the blood cell imaging picture is started;

[0093] Error risk analysis is used to identify error areas in blood cell imaging images. These areas are areas where the risk of blood cell analysis errors is high due to blood cell overlap. By rotating the converter to replace the high-power objective lens and repeatedly adjusting the focus until a clear image is observed under the high-power lens, a high-definition image of the area corresponding to the error area is obtained.

[0094] The blood cell imaging image and the regional high-definition image are respectively input into the convolutional neural network. After the blood cell characteristic information is output, it is summarized and marked on the blood cell imaging image to obtain the corrected blood cell characteristic information. The corrected blood cell characteristic information is input into the disease auxiliary diagnosis model, and the type of disease present in the smear and the corresponding probability are output.

[0095] It is necessary to further explain in the embodiment of the present invention that the error area to be analyzed is obtained by dividing the blood cell imaging image into a number of regions of equal area, with k regions being provided, and s being used to represent the region number; obtaining the overlap degree and the total area and clarity of the connected regions of each region, which are respectively recorded as co_s, as_s, and xs_s; and using the formula The error risk coefficient wf of each area is calculated, where α represents the influence coefficient of the product of clarity and area, β represents the influence coefficient of overlap, and Fnorm(·) represents the linear normalization function, which is used to limit the value of overlap to the range of 0 to 1. Based on the preset error threshold, the error area to be analyzed is screened.

[0096] It needs to be further explained in the embodiment of the present invention that the method for obtaining the corrected blood cell characteristic information is: replacing the blood cell characteristic information at the corresponding position in the blood cell imaging picture with the blood cell characteristic information corresponding to the regional high-definition image, and outputting the corrected blood cell characteristic information.

[0097] Summary: In order to avoid errors in blood cell analysis caused by a small amount of overlapping blood cells in the smear, the error area to be analyzed in the blood cell imaging picture is identified, and a high-definition image of the region corresponding to the error area to be analyzed is obtained; the blood cell imaging picture and the high-definition image of the region are respectively input into the convolutional neural network, and the blood cell characteristic information is output and summarized to obtain the corrected blood cell characteristic information; this can avoid inaccurate analysis results caused by errors; and also avoid high-definition analysis of smears, which results in too large data size and occupies too many analysis resources.

[0098] Example 3

[0099] See Figure 2 The present invention provides a blood cell analysis system based on microscopic imaging, comprising:

[0100] The blood cell imaging picture acquisition module obtains several microscopic images of the smear by rotating the coarse focusing screw and the fine focusing screw of the microscope, and obtains the blood cell imaging picture corresponding to each smear;

[0101] The smear quality analysis module is used to obtain the smear quality evaluation index TQ of the blood cell imaging image; obtain the blood cell overlap CO of the blood cell imaging image, and calculate the smear uniformity coefficient TO based on the position coordinates of the connected area; and jointly analyze the smear uniformity coefficient and the blood cell overlap to obtain the smear quality evaluation index TQ;

[0102] The smear quality judgment module is used to judge the relationship between the smear quality assessment index and the threshold; when the smear quality assessment index is not lower than the threshold, the blood cell imaging picture qualified instruction is output; based on the blood cell imaging picture qualified instruction, the imaging picture feature analysis instruction is started;

[0103] The blood cell imaging picture feature analysis module is used to obtain the blood cell feature information corresponding to the smear, input the blood cell imaging picture into the convolutional neural network, and output the blood cell feature information;

[0104] The disease risk auxiliary analysis module is used to obtain the correlation between disease and blood cell characteristic information through correlation analysis, match the disease risk probability for each smear based on the correlation, and build a disease auxiliary diagnosis model.

[0105] It needs to be further explained in the embodiment of the present invention that the system further includes:

[0106] The blood cell characteristic information correction module is used to correct the blood cell characteristic information and use the corrected blood cell characteristic information to assist in disease risk assessment; it includes:

[0107] Based on the qualified instruction of the blood cell imaging picture, an error risk analysis instruction for the blood cell imaging picture is started;

[0108] Error risk analysis is used to identify error areas in blood cell imaging images. These areas are areas where the risk of blood cell analysis errors is high due to blood cell overlap. By rotating the converter to replace the high-power objective lens and repeatedly adjusting the focus until a clear image is observed under the high-power lens, a high-definition image of the area corresponding to the error area is obtained.

[0109] The blood cell imaging image and the regional high-definition image are respectively input into the convolutional neural network. After the blood cell characteristic information is output, it is summarized and marked on the blood cell imaging image to obtain the corrected blood cell characteristic information. The corrected blood cell characteristic information is input into the disease auxiliary diagnosis model, and the type of disease present in the smear and the corresponding probability are output.

[0110] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A blood cell analysis method based on microscopic imaging, characterized in that: The following steps are involved: Step 1: Obtain several microscopic images of the smear by rotating the coarse focusing screw and the fine focusing screw, and output a blood cell imaging image; Step 2: Obtain the blood cell overlap CO of the blood cell imaging image, and calculate the smear uniformity coefficient TO based on the position coordinates of the connected area; jointly analyze the smear uniformity coefficient and the blood cell overlap to obtain the smear quality assessment index TQ; calculate the smear quality assessment index TQ using the formula TQ = w1*TO + W2*(1-CO), where w1 represents the influence coefficient of the smear uniformity coefficient and w2 represents the influence coefficient of the overlap; Step 3: Determine the relationship between the smear quality assessment index and the threshold; Step 4: Blood cell imaging image feature analysis: Input the blood cell imaging image into the convolutional neural network, output the blood cell feature information, and mark the blood cell imaging image; Step 5. Disease risk auxiliary analysis: The correlation between the disease and blood cell characteristic information is obtained through correlation analysis. The disease risk probability is matched for each smear based on the correlation, and a disease auxiliary diagnosis model is built.

2. The blood cell analysis method based on microscopic imaging according to claim 1, characterized in that: The blood cell overlap degree of the blood cell imaging picture is obtained as follows: Step S01: converting the blood cell imaging picture into a grayscale image, and using Gaussian filtering and smoothing technology to remove image noise; converting the blood cell imaging picture into a binary image; Step S02, morphological analysis: using morphological operations and connected region labeling algorithms to identify each blood cell or blood cell group in the image and mark connected regions; there are m connected regions, and j represents the number of connected regions; Step S03: Get the area of ​​each connected region, record the area of ​​the j-th connected region as ASj, and use the formula The total blood cell area Ato was calculated; Step S04, identifying overlapping regions and calculating the area of ​​overlapping regions: setting the distance between the centers of blood cells to be less than the overlap judgment threshold, the blood cells are considered to be overlapping and the overlapping regions are marked; calculating the area of ​​overlapping regions through morphological operations, recording the overlapping area of ​​the j-th connected region as ACj, and using the formula The overlapping area Atv is calculated; Step S05: By formula The blood cell overlap degree CO of the blood cell imaging image is calculated.

3. The blood cell analysis method based on microscopic imaging according to claim 2, characterized in that: The smear uniformity coefficient is obtained by: obtaining the two-dimensional position coordinates of each connected area on the smear through image recognition technology; dividing the smear into multiple grids of equal area, calculating the number of connected areas in each grid; calculating the density of the connected areas in each grid, and expressing the smear uniformity coefficient TO as the inverse of the variance of the connected area density.

4. The blood cell analysis method based on microscopic imaging according to claim 3, characterized in that: When the smear quality assessment index is lower than the threshold, an early warning is issued to prompt the smear to be replaced; when the smear quality assessment index is not lower than the threshold, a blood cell imaging picture qualified instruction is output; based on the blood cell imaging picture qualified instruction, the imaging picture feature analysis instruction is started.

5. The blood cell analysis method based on microscopic imaging according to claim 1, characterized in that: The convolutional neural network is a trained blood cell type recognition model. The process of building the blood cell type recognition model includes the following steps: Step S11: obtaining a blood cell imaging image of a smear, manually marking the actual characteristic information of each blood cell, and obtaining a sample data set; Step S12: using the blood cell imaging image as input to a blood cell type recognition model, and outputting predicted feature information of each blood cell; Step S13, model initialization: defining the initial parameters of deep learning, including weight parameters, bias parameters, and activation functions between neural networks; Step S14: Divide the sample data set into model training samples and test samples according to the proportion, input the training samples into the convolutional neural network, update the inter-neural network weights, bias parameters and activation functions of the neural network through forward propagation and backward propagation of the neural network, calculate the loss function, and set the loss function based on the similarity between the predicted feature information and the actual feature information; Step S15: If the loss value does not meet the preset threshold requirement, the weight parameters and bias parameters need to be updated through the back propagation algorithm; the back propagation algorithm calculates the gradient of the loss function with respect to each weight and bias, that is, the partial derivative of the loss value with respect to these parameters, and updates the value of each parameter based on the gradient information; Step S16, iterative training: Repeat the steps of forward propagation, loss calculation, backpropagation, and parameter update until the loss value meets the preset threshold requirement or reaches a predetermined number of iterations, thereby obtaining a trained blood cell type recognition model; verify the accuracy and precision of the model through test samples, and deploy the trained blood cell type recognition model into the application.

6. The blood cell analysis method based on microscopic imaging according to claim 1, characterized in that: The construction of the disease-assisted diagnosis model includes the following steps: Step S21, data collection and preprocessing: obtaining blood cell characteristic information corresponding to each disease type, checking and processing missing values, abnormal values ​​and duplicate records, and outputting a sample set; Step S22, data partitioning: Divide the sample set into a training set, a validation set, and a test set; the training set is used to train the decision tree, allowing the decision tree to learn the mapping relationship from blood cell feature information to the target variable; the validation set is used to adjust the parameters of the decision tree, including the number of trees in the random forest, the maximum depth of the tree, and the maximum number of features considered when splitting a node; the test set is used to evaluate the performance indicators of the decision tree; Step S23, constructing a random forest: a decision tree algorithm is selected to construct a single decision tree; when constructing each tree, a subset is extracted from the training set through sampling with replacement as the training data for the tree; when splitting each node of the tree, only the randomly selected feature subset is considered; the above process is repeated to construct multiple decision trees to form a random forest; the prediction results of multiple trees are integrated using majority voting or regression method; Step S24: Use the test set to evaluate the performance of the random forest. The obtained evaluation indicators include accuracy, recall rate, F1 score, and area under the ROC curve; adjust the parameters of the random forest through the cross-validation method until the evaluation indicator requirements are met, and output the disease auxiliary diagnosis model.

7. The blood cell analysis method based on microscopic imaging according to claim 4, characterized in that: The blood cell analysis method further comprises: Based on the qualified instruction of the blood cell imaging picture, an error risk analysis instruction for the blood cell imaging picture is started; Error risk analysis is used to identify error areas in blood cell imaging images. These areas are areas where the risk of blood cell analysis errors is high due to blood cell overlap. By rotating the converter to replace the high-power objective lens and repeatedly adjusting the focus until a clear image is observed under the high-power lens, a high-definition image of the area corresponding to the error area is obtained. The blood cell imaging image and the regional high-definition image are respectively input into the convolutional neural network. After the blood cell characteristic information is output, it is summarized and marked on the blood cell imaging image to obtain the corrected blood cell characteristic information. The corrected blood cell characteristic information is input into the disease auxiliary diagnosis model, and the type of disease present in the smear and the corresponding probability are output.

8. The blood cell analysis method based on microscopic imaging according to claim 7, characterized in that: The error area to be analyzed is obtained by dividing the blood cell imaging image into several regions of equal area, with k regions and s representing the region number; obtaining the overlap of each region and the total area and clarity of the connected regions, which are recorded as co_s, as_s, and xs_s respectively; and using the formula The error risk coefficient wf of each area is calculated, where α represents the influence coefficient of the product of clarity and area, β represents the influence coefficient of overlap, and Fnorm(·) represents the linear normalization function, which is used to limit the value of overlap to the range of 0 to 1. Based on the preset error threshold, the error area to be analyzed is screened.

9. A blood cell analysis system based on microscopic imaging, characterized in that: include: The blood cell imaging picture acquisition module obtains several microscopic images of the smear by rotating the coarse focusing screw and the fine focusing screw of the microscope, and obtains the blood cell imaging picture corresponding to each smear; The smear quality analysis module is used to obtain the smear quality evaluation index TQ of the blood cell imaging image; obtain the blood cell overlap CO of the blood cell imaging image, and calculate the smear uniformity coefficient TO based on the position coordinates of the connected area; The smear quality assessment index TQ was obtained by jointly analyzing the smear uniformity coefficient and blood cell overlap; The smear quality judgment module is used to judge the relationship between the smear quality assessment index and the threshold; when the smear quality assessment index is not lower than the threshold, the blood cell imaging picture qualified instruction is output; based on the blood cell imaging picture qualified instruction, the imaging picture feature analysis instruction is started; The blood cell imaging picture feature analysis module is used to obtain the blood cell feature information corresponding to the smear, input the blood cell imaging picture into the convolutional neural network, and output the blood cell feature information; The disease risk auxiliary analysis module is used to obtain the correlation between disease and blood cell characteristic information through correlation analysis, match the disease risk probability for each smear based on the correlation, and build a disease auxiliary diagnosis model.

Images