Auxiliary detection method and system for blood cell morphology

Abstract

The present invention relates to the field of image analysis technology, and specifically to an auxiliary detection method and system for blood cell morphology. The present invention first analyzes and obtains the differentiation performance coefficient of each blood cell in the blood cell grayscale image; then obtains the matching difference parameters between the key points in different blood cells; further obtains the differential differentiation coefficients between different blood cells; thereby classifying all blood cells, and determining the differentiation stage of the blood cells in the bone marrow sample to be analyzed based on the classification results. The present invention quantitatively analyzes the differentiation performance coefficient of each blood cell according to the different morphological structures of blood cells in each differentiation stage; then, based on the feature matching information and grayscale information of the key points in the blood cells, comprehensively evaluates the differential differentiation coefficients between two blood cells, thereby classifying the blood cells, providing a comparative basis for the subsequent detection and classification of blood cells in the bone marrow sample to be analyzed, thereby improving the detection and evaluation effect and efficiency of the differentiation stage of blood cells.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and in particular to an auxiliary detection method and system for blood cell morphology. Background Art

[0002] Blood cell morphology refers to the discipline that observes and analyzes the morphology, structure, and number of various blood cells in the blood. Based on blood cell morphology, the number and differentiation stage of blood cells can be detected and analyzed, which is of great value for the diagnosis, treatment, and prevention of various blood diseases. The detection and analysis of blood cells is usually based on bone marrow sample smears to manually analyze each blood cell. The process is cumbersome and easily affected by subjective factors, resulting in poor blood cell detection results. Therefore, the use of deep learning algorithms to detect and analyze the number and differentiation stage of blood cells has gradually become a mainstream method.

[0003] In the prior art, a large number of images of blood cells at different differentiation stages are usually manually annotated to construct a training sample set, and a blood cell classification model is trained based on a deep learning algorithm, so that the trained blood cell classification model can be used to evaluate the differentiation stage of each blood cell in the current bone marrow sample; however, human blood cells have a wide variety of morphologies in different differentiation processes, and their spatial directionality also varies greatly. The process of collecting and annotating blood cell samples and training models is complicated, and the recognition effect is easily affected by the training set and training hyperparameters, which leads to poor detection and evaluation of the differentiation stages of blood cells. Summary of the invention

[0004] In order to solve the technical problem that the existing technology has poor detection and evaluation effect on the differentiation stage of blood cells, the purpose of the present invention is to provide an auxiliary detection method and system for blood cell morphology. The technical solution adopted is as follows:

[0005] The present invention proposes an auxiliary detection method for blood cell morphology, the method comprising:

[0006] Acquire blood cell grayscale images of all bone marrow samples, and acquire all key points and all closed edges in each blood cell in each of the blood cell grayscale images;

[0007] In each of the blood cells, according to the local grayscale information of all the key points in the area enclosed by each closed edge, combined with the shape information of each closed edge, the differentiation performance coefficient of each of the blood cells is obtained; according to the local gradient direction difference between the key points in different blood cells and the position distribution difference in the corresponding blood cells, combined with the local grayscale difference, the matching difference parameters between the corresponding key points are obtained;

[0008] Between different blood cells, based on the matching difference parameters of all different key points and combined with the differences in the differentiation performance coefficients between the corresponding blood cells, the differential differentiation coefficients between the corresponding blood cells are obtained; all the blood cells are classified according to the differential differentiation coefficients, and the differentiation stage of the blood cells of the bone marrow sample to be analyzed is determined according to the classification results.

[0009] Furthermore, the method for obtaining the differentiation performance coefficient includes:

[0010] Constructing a preset first neighborhood with each of the key points as the center, and obtaining a local grayscale coefficient according to the grayscale difference between each of the key points and all other pixel points in the corresponding preset first neighborhood;

[0011] The area enclosed by each closed edge is taken as a cell structure area, and it is determined whether each cell structure area contains other closed edges; when not contained, the differentiation sub-parameters of the cell structure area are obtained according to the local grayscale coefficients of all the key points in the cell structure area; when contained, the areas enclosed by other closed edges in each cell structure area are taken as sub-structure areas of the cell structure area, and the differentiation sub-parameters of each cell structure area are obtained according to the area size of each sub-structure area and the local grayscale coefficients of all the key points therein;

[0012] The differentiation sub-parameters of all the cell structure regions in each of the blood cells are integrated to obtain the differentiation performance coefficient of each of the blood cells.

[0013] Furthermore, the method for obtaining the differentiation sub-parameters includes:

[0014] When it is not included, taking the coefficient difference of the local grayscale coefficients of every two key points in the cell structure region as the first coefficient difference, and taking the mean of all the first coefficient differences as the differentiation sub-parameter of the cell structure region;

[0015] When included, the coefficient difference of the local grayscale coefficients of every two key points in each substructure area is taken as the second coefficient difference, and the average of all the second coefficient differences is taken as the structural diversity coefficient of each substructure area; the area proportion of each substructure area in the corresponding cell structure area is obtained; the structural diversity coefficient of the corresponding substructure area is weightedly averaged using the area proportion, and the weighted average result is taken as the differentiation sub-parameter of the corresponding cell structure area.

[0016] Furthermore, the method for obtaining the matching difference parameter includes:

[0017] A polar coordinate system is constructed with the geometric center of each blood cell as the origin; and distance difference parameters corresponding to different key points are obtained according to the differences between the polar diameters of different key points among different blood cells;

[0018] Constructing a preset second neighborhood with each key point as the center; between different blood cells, obtaining local feature difference parameters corresponding to different key points according to the gradient direction information of all pixel points in the preset second neighborhood corresponding to different key points, combined with the difference of the local grayscale coefficients corresponding to different key points;

[0019] According to the distance difference parameters and the local feature difference parameters of different key points, matching difference parameters corresponding to different key points are obtained; the distance difference parameter and the local feature difference parameter are both positively correlated with the matching difference parameter.

[0020] Furthermore, the method for obtaining the local feature difference parameter includes:

[0021] According to the gradient direction histogram of all neighboring pixel points of each key point in the preset second neighborhood, the main direction and all auxiliary directions of each key point are obtained; between different key points, the main directions are aligned to obtain the angle difference between the auxiliary directions, and when the angle difference takes the minimum value, the minimum value is used as the local direction difference parameter;

[0022] Using the difference of the local grayscale coefficients between different key points as a local grayscale difference parameter;

[0023] According to the local direction difference parameter and the local grayscale difference parameter, local feature difference parameters between different key points are obtained; the local direction difference parameter and the local grayscale difference parameter are both positively correlated with the local feature difference parameter.

[0024] Furthermore, the method for obtaining the difference differentiation coefficient includes:

[0025] Obtaining a differentiation difference weight according to the coefficient difference of the differentiation performance coefficients between different blood cells; the coefficient difference is positively correlated with the differentiation difference weight;

[0026] Between different blood cells, the matching difference parameters of all different key points are integrated to obtain cell matching difference parameters;

[0027] The cell matching difference parameter is weighted using the differentiation difference weight, and the weighted result is used as the difference differentiation coefficient between different blood cells.

[0028] Furthermore, the method for classifying all the blood cells according to the differential differentiation coefficient includes:

[0029] The differential differentiation coefficients between different blood cells are used as a metric distance, and all the blood cells are classified based on a distance clustering algorithm.

[0030] Furthermore, the method for determining the differentiation stage of blood cells in the bone marrow sample to be analyzed according to the classification result includes:

[0031] The cluster center and the corresponding differentiation stage of each classification result are obtained, and any blood cell in the bone marrow sample to be analyzed is taken as the target blood cell; the differential differentiation coefficient between the target blood cell and each of the cluster centers is obtained, and when the differential differentiation coefficient is the smallest, it is determined that the differentiation stage of the target blood cell is consistent with the classification result corresponding to the cluster center.

[0032] Furthermore, the obtaining of local feature difference parameters between different key points according to the local direction difference parameter and the local grayscale difference parameter includes:

[0033] The Euclidean norm of the local direction difference parameter and the local grayscale difference parameter is used as the local feature difference parameter corresponding to different key points.

[0034] The present invention also proposes an auxiliary detection system for blood cell morphology, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the auxiliary detection method for blood cell morphology are implemented.

[0035] The present invention has the following beneficial effects:

[0036] The embodiment of the present invention obtains the grayscale images of blood cells of all bone marrow samples, and obtains all key points and all closed edges in each blood cell in each blood cell grayscale image, so as to prepare for the subsequent blood cell matching and classification; in each blood cell, according to the local grayscale information of all key points in the area enclosed by each closed edge, combined with the shape information of each closed edge, the differentiation performance coefficient of each blood cell is obtained, and the differentiation state performance of each blood cell at the differentiation stage is quantified; according to the local gradient direction difference between the key points in different blood cells and the position distribution difference in the corresponding blood cells, combined with the local grayscale difference, the matching difference parameters between the corresponding key points are obtained; between different blood cells, according to the matching difference parameters of all different key points, combined with the difference in the differentiation performance coefficients between the corresponding blood cells, the differential differentiation coefficients between the corresponding blood cells are obtained, and the larger the differential differentiation coefficient, the lower the possibility that the two blood cells are at the same differentiation stage; all blood cells are classified according to the differential differentiation coefficient, and the differentiation stage of the blood cells of the bone marrow sample to be analyzed is determined according to the classification result. The present invention firstly quantitatively analyzes the differentiation performance coefficient of each blood cell according to the different morphological structures of blood cells in each differentiation stage; then, based on the feature matching information and grayscale information of key points in the blood cells, comprehensively evaluates the differential differentiation coefficient between two blood cells, thereby classifying all blood cells, providing a comparative basis for the subsequent detection and classification of blood cells in the spinal cord samples to be analyzed, thereby improving the detection and evaluation effect and efficiency of the differentiation stage of blood cells. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A flow chart of an auxiliary detection method for blood cell morphology provided by one embodiment of the present invention;

[0039] Figure 2 A flowchart of a method for obtaining a differentiation performance coefficient provided by an embodiment of the present invention;

[0040] Figure 3 A flowchart of a method for obtaining matching difference parameters provided by an embodiment of the present invention;

[0041] Figure 4 A gradient direction histogram of a key point provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the auxiliary detection method and system for blood cell morphology proposed by the present invention, its specific implementation, structure, characteristics and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0043] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0044] The specific scheme of the auxiliary detection method and system for blood cell morphology provided by the present invention is described in detail below with reference to the accompanying drawings.

[0045] See also Figure 1 , which shows a flow chart of an auxiliary detection method for blood cell morphology provided by an embodiment of the present invention, specifically comprising:

[0046] Step S1, obtaining blood cell grayscale images of all bone marrow samples, and obtaining all key points and all closed edges in each blood cell in each blood cell grayscale image.

[0047] In one embodiment of the present invention, firstly, grayscale images of blood cells of all bone marrow samples are obtained in the database; wherein the method for obtaining the grayscale images of blood cells is roughly as follows: the collected bone marrow samples are smeared on a glass slide to form a sample smear; the smear is stained using Wright's staining method to enhance the morphology and structural characteristics of the cells for easy observation and analysis under a microscope; the processed smear is placed under an electron microscope for imaging, and the obtained color image is transmitted to a computer for Gaussian denoising, and further grayscale processing is performed to obtain a grayscale image of blood cells of each bone marrow sample; this is a well-known technology and will not be described in detail.

[0048] The blood cells in the bone marrow sample include red blood cells, white blood cells, platelets and a series of cells in the corresponding differentiation stages. For example, the cells in various differentiation stages of red blood cells include multipotent hematopoietic stem cells, myeloid stem cells, erythroid progenitor cells, proerythroid cells, proerythroid cells, proerythroid cells, meserythroid cells, metaerythroid cells, reticulocytes and the final red blood cells; white blood cells and platelets also include cells in various differentiation stages, which are not elaborated here.

[0049] To facilitate the subsequent matching and classification of blood cells, an embodiment of the present invention further uses the trained semantic segmentation model to segment and obtain all blood cells in each blood cell grayscale image; then uses the Scale-Invariant Feature Transform (SIFT) algorithm to detect all key points in each blood cell, so as to perform subsequent feature matching between blood cells; at the same time, considering that blood cells include cell membranes, cytoplasm and various organelles, and the cytoplasm and organelles are all wrapped by cell membranes, the entire blood cell can be regarded as a multi-level wrapping structure, each structure can be regarded as a closed area, and there are obvious differences in the boundaries between different cell structures, so canny edge detection is further performed on the blood cells, and all detected edges are subjected to edge tracking detection to obtain all closed edges in each blood cell.

[0050] It should be noted that the training and application of semantic segmentation models, and the detection of key points and closed edges are all existing technologies well known to those skilled in the art; in other embodiments of the present invention, implementers may also adopt other methods such as the Hessian corner detection algorithm to obtain key points, and use the Sobel operator to obtain edges, etc., which will not be repeated here.

[0051] Step S2, in each blood cell, according to the local grayscale information of all key points in the area enclosed by each closed edge, combined with the shape information of each closed edge, obtain the differentiation performance coefficient of each blood cell; according to the local gradient direction differences between the key points in different blood cells and the position distribution differences in the corresponding blood cells, combined with the local grayscale differences, obtain the matching difference parameters between the corresponding key points.

[0052] Taking into account the differences in the structure and morphological characteristics of blood cells at different differentiation stages, which are specifically reflected in the differences in the grayscale expression and shape size of the cell structure in different blood cells; and considering that based on the above structural analysis of blood cells, a closed edge can be regarded as a cell structure of a blood cell, and the differentiation stage it is in can be determined by analyzing the grayscale expression and morphological expression of each cell structure; therefore, in each blood cell, the embodiment of the present invention obtains the differentiation expression coefficient of each blood cell based on the local grayscale information of all key points in the area enclosed by each closed edge, combined with the shape information of each closed edge; the differentiation expression coefficient roughly quantifies the differentiation state expression of each blood cell at the differentiation stage.

[0053] Preferably, in one embodiment of the present invention, the method for obtaining the differentiation performance coefficient includes:

[0054] See also Figure 2 , which shows a flow chart of a method for obtaining a differentiation performance coefficient provided by an embodiment of the present invention, including:

[0055] Step S201, constructing a preset first neighborhood with each key point as the center, and obtaining a local grayscale coefficient according to the grayscale difference between each key point and all other pixels in the corresponding preset first neighborhood.

[0056] Considering that the color of the cytoplasm in blood cells will change as the differentiation state of blood cells changes, and the key points are important pixel points for subsequent feature matching of blood cells, the differentiation state performance can be preliminarily determined based on the local grayscale information of the key points, and then the differentiation state performance of each blood cell can be evaluated in combination with the morphological structure.

[0057] As an example, firstly, a preset first neighborhood with a size of 5×5 is constructed with each key point as the center, and then the absolute value of the difference between the grayscale value of the key point and each other pixel point in the preset first neighborhood is calculated, and all the absolute values ​​of the difference are averaged to obtain the local grayscale coefficient of each key point; the local grayscale coefficient reflects the local grayscale performance characteristics of each key point, and can preliminarily provide relevant evaluation information from the color angle for the differentiation status of blood cells.

[0058] In other examples, implementers may also set the neighborhood size by themselves, which will not be described in detail here.

[0059] Step S202, taking the area enclosed by each closed edge as a cell structure area, and determining whether each cell structure area contains other closed edges; if not, obtaining the differentiation sub-parameters of the cell structure area according to the local grayscale coefficients of all key points in the cell structure area; if contained, taking the areas enclosed by other closed edges in each cell structure area as sub-structure areas of the cell structure area, and obtaining the differentiation sub-parameters of each cell structure area according to the area size of each sub-structure area and the local grayscale coefficients of all key points therein.

[0060] In one embodiment of the present invention, considering that a closed edge may represent an organelle or a substructure inside an organelle in a blood cell, such as a cell membrane, a cell nucleus, a mitochondria, and a nucleolus in the cell nucleus, etc., as an example, firstly, the area enclosed by each closed edge is taken as a cell structure area, and it is determined whether each cell structure area contains other closed edges; and further, the contribution of each cell structure area in the blood cell to the differentiation state of the blood cell is evaluated;

[0061] For example, if the closed edge corresponds to the cell membrane, it may contain other closed edges corresponding to mitochondria, cell nucleus and its nucleolus, then the area enclosed by other closed edges corresponding to mitochondria, cell nucleus and its nucleolus can be used as the substructure area corresponding to the cell membrane; if the closed edge corresponds to the cell nucleus, it contains the closed edge corresponding to the nucleolus, then the closed edge corresponding to the nucleolus can be used as the substructure area corresponding to the cell nucleus; but the closed edge may also correspond to a smallest structure in the smallest organelle, such as the nucleolus, in which the closed edge may not be detected, then the nucleolus can be used as a cell structure area alone, which does not contain other substructure areas;

[0062] Considering that each cell structure region may contain several substructure regions, and each substructure region may contain multiple key points, the local grayscale coefficients of all key points can be combined to evaluate the local grayscale information of the current substructure region. If the local grayscale difference of all key points is smaller, it means that the substructure region is more likely to be a single organelle or cell structure. At the same time, with the change of differentiation stage, the number and size of various organelles such as nuclei inside the cell gradually decrease or disappear. The larger the area of ​​the substructure region in the corresponding cell structure region and the more substructure regions there are, the more likely it is to correspond to the early stage of differentiation.

[0063] Therefore, in this example, when each cell structure region does not contain other closed edges, the differentiation sub-parameters of the cell structure region are obtained according to the local grayscale coefficients of all key points in the cell structure region; when each cell structure region contains other closed edges, the area enclosed by the other closed edges in each cell structure region is taken as the sub-structure region of the cell structure region, and the differentiation sub-parameters of each cell structure region are obtained according to the area size of each sub-structure region and the local grayscale coefficients of all key points therein.

[0064] In a preferred embodiment of the present invention, the method for obtaining the differentiation sub-parameters includes:

[0065] When it is not included, the coefficient difference of the local grayscale coefficient of every two key points in the cell structure area is taken as the first coefficient difference, and the mean of all first coefficient differences is taken as the differentiation sub-parameter of the cell structure area;

[0066] When included, the coefficient difference of the local grayscale coefficients of every two key points in each substructure area is taken as the second coefficient difference, and the average of all second coefficient differences is taken as the structural diversity coefficient of each substructure area; the area proportion of each substructure area in the corresponding cell structure area is obtained; the structural diversity coefficient of the corresponding substructure area is weighted and averaged using the area proportion, and the weighted average result is taken as the differentiation sub-parameter of the corresponding cell structure area.

[0067] As an example, when included, the differentiation sub-parameter is calculated as:

[0068] ;in, For the of blood cells Differentiation parameters of each cell structure area; For the of blood cells The serial number of the substructure area in each cell structure area; For the of blood cells The total number of substructure regions in each cell structure region; For the of blood cells In the cellular structural area The structural diversity coefficient of each substructure region; For the of blood cells In the cellular structural area The area of ​​each substructure region; For the of blood cells The area of ​​the cell structure region; For area ratio.

[0069] Among them, the smaller the structural diversity coefficient, the greater the possibility that the substructure region represents the same organelle. At the same time, if the corresponding area ratio is smaller, it means that the cell structure region where the substructure region is located is more likely to be the cell structure in the blood cell in the late stage of differentiation. The differentiation state level of the overall cell structure region is obtained by averaging.

[0070] Step S203, integrating the differentiation sub-parameters of all cell structure regions in each blood cell to obtain the differentiation performance coefficient of each blood cell.

[0071] As an example, the differentiation sub-parameters of all cell structure regions in each blood cell are averaged to obtain the differentiation performance coefficient of each blood cell; the differentiation performance coefficient quantifies the level of differentiation state of each blood cell, and the smaller the differentiation performance coefficient, the more the blood cell is in the early stage of differentiation.

[0072] At this point, the differentiation expression coefficient of each blood cell was obtained.

[0073] Considering that blood cells are of many types and have different morphologies, the SIFT algorithm can be used to perform feature matching on blood cells to classify cells at the same differentiation stage, thereby providing a sample basis for the detection and identification of blood cells in newly collected bone marrow samples. By directly comparing the blood cells in the newly collected bone marrow samples with the sample library constructed from all classified blood cells, their differentiation stage and quantity can be evaluated, thereby improving the efficiency of blood cell detection.

[0074] It is also considered that when determining key points based on the SIFT algorithm, each key point has three pieces of information: position, scale and direction; since it is feature matching of blood cells, but the acquisition of blood cell grayscale images is a standard experimental process, the scale of blood cells in all blood cell grayscale images is consistent, and the scale of all key points therein is also consistent; but the orientation of blood cells in blood cell grayscale images is not fixed, and the position distribution of each organelle in the blood cells is relatively stable, then the position distribution of the matched key points in the corresponding blood cells should also be relatively similar, and at the same time, the local gradient direction characteristics and local grayscale characteristics of the matched key points should also be similar.

[0075] Therefore, in the embodiment of the present invention, the matching difference parameters between the corresponding key points are obtained according to the difference in the local gradient direction of different key points and the difference in the position distribution in the corresponding blood cells, combined with the local grayscale difference of the corresponding key points. The matching difference parameter determines the possibility of feature matching by analyzing the difference in local feature information between a pair of key points to be matched in different blood cells, so as to facilitate the subsequent comprehensive evaluation of the matching possibility of all key points in different blood cells to assess whether the corresponding blood cells are at the same differentiation stage. The larger the matching difference parameter, the less likely it is that the two blood cells are at the same differentiation stage.

[0076] Preferably, in one embodiment of the present invention, the method for obtaining the matching difference parameter includes:

[0077] See also Figure 3 , which shows a flow chart of a method for obtaining matching difference parameters provided by an embodiment of the present invention, including:

[0078] Step S301, constructing a polar coordinate system with the geometric center of each blood cell as the origin; obtaining distance difference parameters corresponding to different key points between different blood cells according to the differences between the polar diameters of different key points.

[0079] As an example, considering that the orientation of each blood cell in the blood cell grayscale image is not fixed, a polar coordinate system is first constructed with the center of mass of each blood cell as the origin to reduce the influence of the traditional Euclidean distance measurement on position similarity; the polar diameter of each key point reflects its relative position in the blood cell, and then the distance difference parameter between the corresponding key points can be obtained through the absolute value of the difference between the polar diameters of different key points. The larger the distance difference parameter, the greater the difference in the position distribution of the key points in the blood cells, and the lower the possibility of subsequent successful matching.

[0080] It should be noted that the construction of polar coordinate system and polar diameter are well-known technologies and will not be described in detail here.

[0081] Step S302, constructing a preset second neighborhood with each key point as the center; between different blood cells, according to the gradient direction difference of all pixel points in the preset second neighborhood of different key points, combined with the difference of local grayscale coefficients corresponding to different key points, obtaining the local feature difference parameters between different key points.

[0082] Considering that the local gradient directions of different key points differ greatly, the matching possibility will also be reduced.

[0083] In a preferred embodiment of the present invention, the method for obtaining the local feature difference parameter includes:

[0084] According to the gradient direction histogram of all neighboring pixels of each key point in the preset second neighborhood, the main direction and all auxiliary directions of each key point are obtained; between different key points, the main directions are aligned to obtain the angle difference between the auxiliary directions, and when the angle difference takes the minimum value, the minimum value is used as the local direction difference parameter;

[0085] The difference of local grayscale coefficients between different key points is used as the local grayscale difference parameter;

[0086] According to the local direction difference parameter and the local grayscale difference parameter, the local feature difference parameters between different key points are obtained; the local direction difference parameter and the local grayscale difference parameter are both positively correlated with the local feature difference parameter.

[0087] As an example, a preset second neighborhood with a size of 9×9 is constructed with each key point as the center, and the gradient direction histogram of all neighboring pixels of each key point is obtained. Figure 4, which shows a gradient direction histogram of a key point provided by an embodiment of the present invention; there are 8 directions in the gradient direction histogram, namely, up, down, left, right and four diagonal directions, that is, starting from 0°, every 45° is a direction interval, totaling eight directions; the implementer can also adjust it according to actual needs, such as taking every 10 degrees as a direction, which is not repeated here; the direction corresponding to the maximum value cylinder in the gradient direction histogram is taken as the main direction of the key point, and the direction corresponding to the cylinder greater than 80% of the maximum value is taken as the auxiliary direction, that is, each key point contains a main direction and several auxiliary directions; by rotating one of the key points, the main directions of the two key points are aligned, and then the angular difference between the auxiliary directions of the two key points is evaluated to obtain the local direction difference coefficient; the larger the local direction difference coefficient, the lower the possibility of successful key point matching.

[0088] Among them, the method for obtaining the local direction difference coefficient specifically includes: for example, between two key points A and B, the auxiliary directions of key point A are a1, a2, and a3, and the auxiliary directions of key point B are b1 and b2. First, among all the auxiliary directions of key point B, search for an auxiliary direction with the smallest absolute value of the angle difference with a1 to obtain the absolute value of the first angle difference; then continue to search for an auxiliary direction with the smallest absolute value of the angle difference with a2 among all the auxiliary directions of key point B to obtain the absolute value of the second angle difference; similarly, obtain the absolute value of the third angle difference, and accumulate all the absolute values ​​of the angle differences to obtain the local direction difference coefficient.

[0089] The absolute value of the difference of the local grayscale coefficients between different key points is further used as the local grayscale difference parameter; the local direction difference parameter and the local grayscale difference parameter are linearly normalized respectively, and the normalized local direction difference parameter and the local grayscale difference parameter are multiplied and combined to obtain the local feature difference parameter between the corresponding key points. It should be noted that when normalizing, if all values ​​are consistent, the normalization results of all values ​​are 1.

[0090] At this point, the local feature difference coefficients of different key points between two different blood cells are obtained.

[0091] Step S303, obtaining matching difference parameters corresponding to different key points according to the distance difference parameters and the local feature difference parameters of different key points; the distance difference parameter and the local feature difference parameter are both positively correlated with the matching difference parameter.

[0092] In a preferred embodiment of the present invention, the Euclidean norm of the local direction difference parameter and the local grayscale difference parameter is used as the local feature difference parameter between the corresponding key points. The calculation formula of the local feature difference parameter is: ;in, In a blood cell The key point is related to the The matching difference parameter between key points; In a blood cell The key point is related to the The distance difference parameter between key points; In a blood cell The key point is related to the The local feature difference parameters between key points.

[0093] In the above formula, when the distance difference parameters and local feature difference parameters of different key points are larger, the matching possibility of two key points is lower, and the corresponding matching difference parameters are larger; other positive correlation merging methods can also be used in other embodiments, which will not be repeated here.

[0094] Step S3, between different blood cells, according to the matching difference parameters of all different key points, combined with the difference in differentiation performance coefficients between corresponding blood cells, obtain the differential differentiation coefficients between corresponding blood cells; classify all blood cells according to the differential differentiation coefficients, and determine the differentiation stage of the blood cells of the bone marrow sample to be analyzed according to the classification results.

[0095] Considering that the matching effect between blood cells requires a comprehensive evaluation of the matching results of all key points therein, and considering that the differentiation performance coefficient of blood cells quantifies the differentiation state performance of each blood cell at the differentiation stage, the differentiation difference and matching difference between two blood cells can be further evaluated based on the difference in the differentiation performance coefficient; the embodiment of the present invention will obtain the differential differentiation coefficient between corresponding blood cells based on the matching difference parameters of all different key points, combined with the difference in the differentiation performance coefficients between corresponding blood cells; the larger the differential differentiation coefficient, the lower the possibility that the two blood cells are blood cells at the same differentiation stage.

[0096] Preferably, in one embodiment of the present invention, the greater the coefficient difference of the differentiation performance coefficient, the greater the differential differentiation coefficient of the two blood cells; and the method for obtaining the differential differentiation coefficient includes:

[0097] According to the coefficient difference of differentiation performance coefficients between different blood cells, the differentiation difference weight is obtained; the coefficient difference is positively correlated with the differentiation difference weight;

[0098] Among different blood cells, the matching difference parameters of all different key points are integrated to obtain the cell matching difference parameters;

[0099] The cell matching difference parameters are weighted using the differentiation difference weights, and the weighted results are used as the differential differentiation coefficients between different blood cells.

[0100] As an example, the absolute value of the difference in differentiation performance coefficients between different blood cells is used as the differentiation difference weight; then the differentiation difference weight is multiplied by the cell matching difference parameter to obtain a weighted result, thereby obtaining the difference differentiation coefficients between different blood cells;

[0101] The method for obtaining the cell matching difference parameters is as follows: in two blood cells, a key point in one blood cell is paired with a key point in the other blood cell, and all the key points are paired without replacement to obtain a matching combination; the key points between the two blood cells are matched to obtain several matching combinations; in each matching combination, the matching difference parameters between the two key points matched in all combinations are accumulated as the cell matching difference result of the matching combination; then a minimum value is screened out from the cell matching difference results of all matching combinations as the cell matching difference parameter between the two blood cells.

[0102] After obtaining the differential differentiation coefficients between different blood cells, all blood cells can be classified according to the differential differentiation coefficients, and then the differentiation stage of the blood cells in the bone marrow sample to be analyzed can be determined according to the classification results.

[0103] Preferably, in one embodiment of the present invention, considering that the larger the differential differentiation coefficient is, the lower the possibility that two blood cells are blood cells at the same differentiation stage is, the differential differentiation coefficient between different blood cells can be used as a metric distance between cluster samples, and all blood cells can be classified based on a distance clustering algorithm.

[0104] As an example, the implementer can use the K-means clustering algorithm to cluster all blood cells in all blood cell grayscale images based on the preset K value, so as to obtain all clusters, and the blood cells in each cluster are blood cells at the same differentiation stage, and the clustering result is the classification result of the blood cells; it should be noted that the preset K value is the number of types of blood cells at each differentiation stage in the spinal cord sample, and the K-means algorithm is a well-known technology and will not be described here.

[0105] Preferably, in one embodiment of the present invention, considering that the classification results obtained based on all spinal cord samples provide database support for the classification detection of blood cells in the subsequently collected spinal cord samples, the implementer can evaluate the differentiation state by analyzing each blood cell in the current spinal cord sample to be analyzed and the blood cell in the classification result; based on this, the method for determining the differentiation stage of the blood cells of the bone marrow sample to be analyzed according to the classification result includes:

[0106] The cluster center and corresponding differentiation stage of each classification result are obtained, and any blood cell in the bone marrow sample to be analyzed is taken as the target blood cell; the differential differentiation coefficient between the target blood cell and each cluster center is obtained. When the differential differentiation coefficient is the smallest, it is determined that the differentiation stage of the classification result to which the target blood cell and the corresponding cluster center belong is consistent.

[0107] It should be noted that the differentiation stage of blood cells corresponding to the cluster center of each classification result is evaluated through manual analysis, and the method for obtaining the differential differentiation coefficient between the target blood cells and each cluster center is consistent with the acquisition method in the above embodiment, which will not be repeated here; when the differential differentiation coefficient is the smallest, it can be determined that the differentiation stage of the target blood cells is consistent with the differentiation stage of the corresponding cluster center.

[0108] After obtaining the differentiation stage of each blood cell in the current spinal cord sample to be analyzed, the blood cells at each differentiation stage can be marked or counted, thereby assisting relevant testing personnel in analyzing and studying the spinal cord samples.

[0109] In summary, the embodiment of the present invention obtains all key points and all closed edges in each blood cell in all bone marrow samples; further analyzes to obtain the differentiation performance coefficient of each blood cell; further analyzes to obtain the matching difference parameters between different key points in different blood cells; further between different blood cells, according to the matching difference parameters of all different key points, combined with the difference in the differentiation performance coefficients between corresponding blood cells, obtains the differential differentiation coefficients between corresponding blood cells; classifies all blood cells according to the differential differentiation coefficients, and determines the differentiation stage of the blood cells in the bone marrow sample to be analyzed according to the classification results. The present invention first quantitatively analyzes the differentiation performance coefficient of each blood cell according to the different morphological structures of blood cells in each differentiation stage; then, based on the feature matching information and grayscale information of the key points in the blood cells, comprehensively evaluates the differential differentiation coefficients between two blood cells, thereby classifying all blood cells, providing a comparative basis for the subsequent detection and classification of blood cells in the spinal cord sample to be analyzed, thereby improving the detection and evaluation effect and efficiency of the differentiation stage of blood cells.

[0110] The present invention also proposes an auxiliary detection system for blood cell morphology. The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, steps S1-S3 of an auxiliary detection method for blood cell morphology described are implemented.

[0111] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0112] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. An auxiliary detection method for blood cell morphology, characterized in that: The method comprises: Acquire blood cell grayscale images of all bone marrow samples, and acquire all key points and all closed edges in each blood cell in each of the blood cell grayscale images; In each of the blood cells, the differentiation performance coefficient of each of the blood cells is obtained according to the local grayscale information of all the key points in the area enclosed by each closed edge, combined with the shape information of each closed edge; the matching difference parameters between the corresponding key points are obtained according to the local gradient direction differences between the key points in different blood cells and the position distribution differences in the corresponding blood cells, combined with the local grayscale differences; Among the different blood cells, according to the matching difference parameters of all the different key points, combined with the differences in the differentiation performance coefficients between the corresponding blood cells, the differential differentiation coefficients between the corresponding blood cells are obtained; according to the differential differentiation coefficients, all the blood cells are classified, and the differentiation stage of the blood cells of the bone marrow sample to be analyzed is determined according to the classification results; The method for obtaining the differentiation performance coefficient includes: Constructing a preset first neighborhood with each of the key points as the center, and obtaining a local grayscale coefficient according to the grayscale difference between each of the key points and all other pixel points in the corresponding preset first neighborhood; The area enclosed by each closed edge is taken as a cell structure area, and it is determined whether each cell structure area contains other closed edges; when not contained, the differentiation sub-parameters of the cell structure area are obtained according to the local grayscale coefficients of all the key points in the cell structure area; when contained, the areas enclosed by other closed edges in each cell structure area are taken as sub-structure areas of the cell structure area, and the differentiation sub-parameters of each cell structure area are obtained according to the area size of each sub-structure area and the local grayscale coefficients of all the key points therein; Integrate the differentiation sub-parameters of all the cell structure regions in each of the blood cells to obtain a differentiation performance coefficient of each of the blood cells; The method for obtaining the differentiation sub-parameters includes: When it is not included, taking the coefficient difference of the local grayscale coefficients of every two key points in the cell structure region as the first coefficient difference, and taking the mean of all the first coefficient differences as the differentiation sub-parameter of the cell structure region; When included, the coefficient difference of the local grayscale coefficients of every two key points in each substructure area is taken as the second coefficient difference, and the average of all the second coefficient differences is taken as the structural diversity coefficient of each substructure area; the area proportion of each substructure area in the corresponding cell structure area is obtained; the structural diversity coefficient of the corresponding substructure area is weighted and averaged using the area proportion, and the weighted average result is taken as the differentiation subparameter of the corresponding cell structure area; The method for obtaining the matching difference parameter includes: A polar coordinate system is constructed with the geometric center of each blood cell as the origin; and distance difference parameters corresponding to different key points are obtained according to the differences between the polar diameters of different key points among different blood cells; Constructing a preset second neighborhood with each key point as the center; between different blood cells, obtaining local feature difference parameters corresponding to different key points according to the gradient direction information of all pixel points in the preset second neighborhood corresponding to different key points, combined with the difference of the local grayscale coefficients corresponding to different key points; According to the distance difference parameters and the local feature difference parameters of different key points, matching difference parameters corresponding to different key points are obtained; the distance difference parameters and the local feature difference parameters are both positively correlated with the matching difference parameters; The method for obtaining the local feature difference parameter includes: According to the gradient direction histogram of all neighboring pixel points of each key point in the preset second neighborhood, the main direction and all auxiliary directions of each key point are obtained; between different key points, the main directions are aligned to obtain the angle difference between the auxiliary directions, and when the angle difference takes the minimum value, the minimum value is used as the local direction difference parameter; Using the difference of the local grayscale coefficients between different key points as a local grayscale difference parameter; According to the local direction difference parameter and the local grayscale difference parameter, local feature difference parameters between different key points are obtained; the local direction difference parameter and the local grayscale difference parameter are both positively correlated with the local feature difference parameter.

2. The auxiliary detection method for blood cell morphology according to claim 1, characterized in that: The method for obtaining the differential differentiation coefficient includes: Obtaining a differentiation difference weight according to the coefficient difference of the differentiation performance coefficients between different blood cells; the coefficient difference is positively correlated with the differentiation difference weight; Between different blood cells, the matching difference parameters of all different key points are integrated to obtain cell matching difference parameters; The cell matching difference parameter is weighted using the differentiation difference weight, and the weighted result is used as the difference differentiation coefficient between different blood cells.

3. The auxiliary detection method for blood cell morphology according to claim 1, characterized in that: The method for classifying all the blood cells according to the differential differentiation coefficient comprises: The differential differentiation coefficients between different blood cells are used as a metric distance, and all the blood cells are classified based on a distance clustering algorithm.

4. The auxiliary detection method for blood cell morphology according to claim 3, characterized in that: The method for determining the differentiation stage of blood cells in the bone marrow sample to be analyzed according to the classification results comprises: The cluster center and the corresponding differentiation stage of each classification result are obtained, and any blood cell in the bone marrow sample to be analyzed is taken as the target blood cell; the differential differentiation coefficient between the target blood cell and each of the cluster centers is obtained, and when the differential differentiation coefficient is the smallest, it is determined that the differentiation stage of the target blood cell is consistent with the classification result corresponding to the cluster center.

5. The auxiliary detection method for blood cell morphology according to claim 1, characterized in that: The obtaining, according to the local direction difference parameter and the local grayscale difference parameter, local feature difference parameters between different key points includes: The Euclidean norm of the local direction difference parameter and the local grayscale difference parameter is used as the local feature difference parameter corresponding to different key points.

6. An auxiliary detection system for blood cell morphology, characterized in that: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of an auxiliary detection method for blood cell morphology as described in any one of claims 1 to 5 are implemented.

Images