An automatic classification detection method for peripheral blood cell morphology

Abstract

The present invention relates to the technical field of cell classification, and particularly relates to a fully automatic classification detection method for peripheral blood cell morphology. The method acquires cell regions in a cell display image of peripheral blood; obtains overlapping cell regions according to the morphology of the cell regions and the curvature of edge pixel points in the cell regions; obtains the exposed edge lines of the covered cells according to the length of each edge line in the overlapping cell regions, the curvature of the edge pixel points on each edge line, and the curvature of the intersection pixel points of each edge line with other edge lines; determines the complete cell edge lines of the covered cells according to the curvature and endpoints of the exposed edge lines, and performs fully automatic classification detection of peripheral blood cell morphology. By acquiring the complete cell edge lines of the covered cells, the present invention determines the morphology of the covered cells, avoids the situation of inaccurate automatic recognition and classification of cells due to cell overlap, and improves the accuracy of fully automatic classification detection of peripheral blood cell morphology.

Description

Technical Field

[0001] The present invention relates to the technical field of cell classification, and particularly to a fully automatic classification detection method for peripheral blood cell morphology. Background Art

[0002] The morphological examination of blood cells is an important diagnostic tool in clinical hematology. By observing the cell morphology in a blood sample under a microscope, it can provide doctors with key information about the patient's health status. This examination plays a crucial role in the early screening and diagnosis of various blood system diseases. Especially in the diagnosis of leukemia, anemia, infections, immune system diseases, and other blood diseases, it can help doctors make more accurate judgments. Traditional morphological examinations of blood cells usually rely on manual microscope observations and have many limitations, such as observer fatigue, subjective bias, slow analysis speed, and poor stability of detection results. To overcome these problems, existing methods use fully automatic classification detection of peripheral blood cell morphology, which uses fully automated equipment to identify and classify various cell types in peripheral blood, analyze cell morphological characteristics, and help doctors quickly and accurately evaluate the patient's health status, especially in the screening and diagnosis of blood diseases, infections, immune diseases, and some other diseases.

[0003] However, in actual situations, during the fully automatic classification detection of peripheral blood cell morphology, there is a problem that the cell distribution is too dense, resulting in cell overlap, and the cell edges cannot be clearly separated. As a result, it is difficult for the automatic classification system to accurately extract the characteristics of each cell, and it is easy to misidentify multiple cells as one cell, making the fully automatic classification detection of peripheral blood cell morphology inaccurate, and thus unable to accurately evaluate the patient's health status. Summary of the Invention

[0004] In order to solve the technical problem that the fully automatic classification detection of peripheral blood cell morphology is inaccurate due to too dense cell distribution resulting in cell overlap, the purpose of the present invention is to provide a fully automatic classification detection method for peripheral blood cell morphology, and the specific technical solution adopted is as follows:

[0005] An embodiment of the present invention provides a fully automatic classification detection method for peripheral blood cell morphology, and this method includes the following steps:

[0006] Obtain the cell regions in the cell display image of peripheral blood;

[0007] Obtain the overlapping cell regions according to the morphology of each cell region and the curvature fluctuation of the edge pixel points in each cell region;

[0008] Obtain the exposed edge lines of the covered cells in the overlapping cell region according to the length of each edge line in the overlapping cell region, the curvature of the edge pixel points on each edge line, and the curvature of the intersection pixel points of each edge line with other edge lines;

[0009] According to the curvature and endpoints of each exposed edge line, obtain the virtual covered edge lines of the covered cells corresponding to each exposed edge line, determine the complete cell edge lines of each covered cell, and perform automatic classification detection of peripheral blood cell morphology.

[0010] Furthermore, the method for obtaining the overlapping cell region is as follows:

[0011] Obtain the shape parameters of each cell region according to the shape of each cell region;

[0012] Obtain the edge undulation degree of each cell region according to the curvature fluctuation of the edge pixel points in each cell region;

[0013] Obtain the overlapping degree of each cell region according to the shape parameters and edge undulation degree of each cell region; wherein, the shape parameter and the overlapping degree are negatively correlated, and the edge undulation degree and the overlapping degree are positively correlated;

[0014] When the overlapping degree is greater than the preset overlapping degree threshold, regard the corresponding cell region as the overlapping cell region.

[0015] Furthermore, the method for obtaining the shape parameter is as follows:

[0016] For any cell region, take the ratio of the area of the cell region to the square of the perimeter of the cell region as the first value;

[0017] Take the product of a specified constant and the first value as the shape parameter of the cell region.

[0018] Furthermore, the method for obtaining the edge undulation degree is as follows:

[0019] For any cell region, obtain the curvature of each edge pixel point in the cell region, and regard them all as reference curvatures;

[0020] Take the difference between the maximum reference curvature and the minimum reference curvature as the first fluctuation degree;

[0021] Take the product of the variance of the reference curvatures and the first fluctuation degree as the edge undulation degree of the cell region.

[0022] Furthermore, the method for obtaining the exposed edge line is as follows:

[0023] Obtain the exposure degree of each edge line in each overlapping cell region based on the length of each edge line, the curvature of the edge pixel points on each edge line, and the curvature of the intersection pixel points of each edge line with other edge lines;

[0024] When the exposure degree is greater than the preset exposure degree threshold, regard the corresponding edge line as an exposed edge line.

[0025] Furthermore, the method for obtaining the exposure degree is as follows:

[0026] For any edge line in any overlapping cell region, obtain the mean value of the curvatures of all edge pixel points on this edge line as the curvature analysis value of this edge line;

[0027] Obtain the mean value of the curvatures of the intersection pixel points of this edge line with other edge lines in this overlapping cell region as the curvature reference value;

[0028] Take the difference between the curvature analysis value and the curvature reference value as the first difference;

[0029] Take the result of normalizing the ratio of the first difference to the number of edge pixel points on this edge line as the exposure degree of this edge line.

[0030] Furthermore, the method for obtaining the exposed edge line also includes:

[0031] Obtain the true exposed edge line according to the gray - scale difference between each edge pixel point on each exposed edge line and its preset neighborhood pixel points.

[0032] Furthermore, the method for obtaining the true exposed edge line is as follows:

[0033] For any edge pixel point on any exposed edge line, regard the straight line perpendicular to the tangent line of this edge pixel point as the target straight line, and regard the preset number of pixel points located on the target straight line and outside the covered cell corresponding to this exposed edge line and closest to this edge pixel point as the preset neighborhood pixel points of this edge pixel point;

[0034] Obtain the difference between the gray - scale value of this edge pixel point and the mean value of the gray - scale values of its preset neighborhood pixel points as the brightness mutation degree of this edge pixel point;

[0035] Take the result of negative correlation and normalization of the sum of the brightness mutation degrees of all edge pixel points on this exposed edge line as the brightness smoothness degree of this exposed edge line;

[0036] When the brightness smoothness degree is greater than the preset brightness smoothness degree threshold, regard the corresponding exposed edge line as the true exposed edge line.

[0037] Further, the method for obtaining the virtual coverage edge line is as follows:

[0038] For any exposed edge line, the pixel points at both ends of the exposed edge line are used as target pixel points;

[0039] For any target pixel point, the eight-neighborhood pixel points of the target pixel point that are not passed by the exposed edge line are used as the adjacent pixel points of the target pixel point;

[0040] By using the Bezier curve technique, the exposed edge line is smoothly connected to each adjacent pixel point of the target pixel point respectively to obtain the reference edge line corresponding to each adjacent pixel point of the target pixel point;

[0041] Obtain the difference in curvature between the reference edge line corresponding to each adjacent pixel point of the target pixel point and the exposed edge line, and all of them are used as the second difference;

[0042] The adjacent pixel points on the reference edge line corresponding to the smallest second difference are used as the first virtual edge pixel points of the exposed edge line;

[0043] By using the Bezier curve technique, the exposed edge line is smoothly connected to the first virtual edge pixel point to obtain the first reference exposed edge line;

[0044] The eight-neighborhood pixel points of the first virtual edge pixel point that are not passed by the first reference exposed edge line are used as the adjacent pixel points of the first virtual edge pixel point;

[0045] By using the Bezier curve technique, the first reference exposed edge line is smoothly connected to each adjacent pixel point of the first virtual edge pixel point respectively to obtain the reference edge line corresponding to each adjacent pixel point of the first virtual edge pixel point;

[0046] Obtain the difference in curvature between the reference edge line corresponding to each adjacent pixel point of the first virtual edge pixel point and the first reference exposed edge line, and all of them are used as the third difference;

[0047] The adjacent pixel points on the reference edge line corresponding to the smallest third difference are used as the second virtual edge pixel points of the exposed edge line;

[0048] And so on, obtain each virtual edge pixel point of the exposed edge line until reaching the other target pixel point of the exposed edge line;

[0049] The virtual edge pixel points obtained with each target pixel point as the starting point are jointly curve-fitted by the least squares method, and the obtained curve is used as the virtual coverage edge line of the covered cell corresponding to the exposed edge line.

[0050] Further, the method for obtaining the cell region is as follows:

[0051] Obtain the cell region in the cell display image through the Canny edge detection algorithm and the connected component algorithm.

[0052] The present invention has the following beneficial effects:

[0053] According to the morphology of each cell region and the curvature fluctuation of the edge pixel points in each cell region, the present invention obtains the overlapping cell region, accurately determines the overlapping cells, which is beneficial to the subsequent accurate and efficient analysis of the covered cells; in order to efficiently obtain the morphology of the covered cells, further according to the length of each edge line in the overlapping cell region, the curvature of the edge pixel points on each edge line, and the curvature of the intersection pixel points of each edge line with other edge lines, the exposed edge line of the covered cells in the overlapping cell region is obtained, and the edge line for the normal display of the covered cells is determined, which is beneficial to the subsequent acquisition of the complete morphology of the covered cells; furthermore, according to the curvature and endpoints of each exposed edge line, the virtual covered edge line corresponding to each exposed edge line of the covered cells is accurately obtained, and the complete cell edge line of each covered cell is determined, effectively solving the problem that the cell morphology is incomplete due to cell overlap, resulting in inaccurate automatic recognition and classification of cells, improving the accuracy of the automatic classification detection of peripheral blood cell morphology, and further accurately evaluating the health status of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0055] Figure 1 It is a schematic flow chart of a method for automatic classification detection of peripheral blood cell morphology provided by an embodiment of the present invention;

[0056] Figure 2 It is a cell display image provided by an embodiment of the present invention;

[0057] Figure 3 It is a flow chart of a method for obtaining an overlapping cell region provided by an embodiment of the present invention;

[0058] Figure 4 It is a partial image of cell overlap provided by an embodiment of the present invention;

[0059] Figure 5 It is a flow chart of a method for obtaining an exposed edge line provided by an embodiment of the present invention;

[0060] Figure 6 Structural diagram of a full-automatic classification detection system for peripheral blood cell morphology provided by an embodiment of the present invention;

[0061] Figure 7 Schematic diagram of a computer device provided by an embodiment of the present invention. Detailed implementation manners

[0062] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following describes in detail a full-automatic classification detection method for peripheral blood cell morphology proposed according to the present invention, its specific implementation manners, structures, features and effects in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0064] The following specifically describes the specific solution of a full-automatic classification detection method for peripheral blood cell morphology provided by the present invention in conjunction with the accompanying drawings.

[0065] Embodiment 1:

[0066] The present invention proposes a full-automatic classification detection method for peripheral blood cell morphology. Please refer to Figure 1 , which shows a schematic flowchart of a full-automatic classification detection method for peripheral blood cell morphology provided by an embodiment of the present invention. The method includes the following steps:

[0067] Step S1: Obtain the cell region in the cell display image of peripheral blood.

[0068] Specifically, in order to analyze the health status of a patient through peripheral blood, in this embodiment, a blood smear is first obtained from the peripheral blood sample of the patient and subjected to a staining process. Common staining methods include Giemsa staining and Wright staining, etc.; then a high-resolution microscope or a scanning microscope is used to observe the blood smear, and at the same time, a cell display image of peripheral blood is obtained. For the convenience of subsequent analysis, the cell display image is grayscale processed as Figure 2 shown. It should be noted that all subsequent cell display images are grayscale cell display images. Among them, Giemsa staining, Wright staining and grayscale processing are all well-known technologies and will not be elaborated further.

[0069] In order to accurately and efficiently analyze peripheral blood cells, in this embodiment, the cell regions in the cell display image are obtained through the Canny edge detection algorithm and the connected component algorithm. Among them, both the Canny edge detection algorithm and the connected component algorithm are well-known technologies and will not be elaborated here.

[0070] Step S2: Obtain the overlapping cell regions according to the morphology of each cell region and the curvature fluctuation of the edge pixel points in each cell region.

[0071] It is known that during the process of fully automatic classification and detection of the morphology of peripheral blood cells, the cells in the cell display image are too densely distributed, resulting in overlapping between cells. As a result, it is difficult for the automatic classification system to accurately extract the features of each cell. The automatic classification system may misidentify multiple overlapping cells as one cell, resulting in incorrect automatic classification, directly affecting the accuracy of peripheral blood cell classification, and thus unable to accurately analyze the health status of the patient. Therefore, this embodiment needs to analyze the overlapping cells to eliminate the inaccurate situation of peripheral blood cell classification. Usually, the shape of a single cell in the blood is round. However, when cells overlap, the shape of the overlapping cells becomes irregular. For example, it may be curved, elongated, or have sharp edges, etc., and the shape significantly deviates from a circle. Therefore, in this embodiment, according to the morphology of each cell region, it is analyzed whether each cell region is an overlapping cell region.

[0072] Considering that the cell morphology is not perfectly close to a circle, only relying on the morphology of each cell region may misjudge some cell regions as overlapping cell regions. Therefore, each cell region needs to be further analyzed. It is known that when cells overlap, there will be obvious fluctuations at the edges of the overlapping parts, that is, the edges are significantly irregular. Therefore, there will be obvious fluctuations in the curvature of the edge pixel points in the cell regions where cells overlap, and the curvature fluctuation of the edge pixel points in the cell region corresponding to a single cell is relatively small. Among them, the method for obtaining the curvature of the edge pixel points is a well-known technology and will not be elaborated here. Therefore, in this embodiment, according to the morphology of each cell region and the curvature fluctuation of the edge pixel points in each cell region, the overlapping cell regions where cells overlap are obtained, which is beneficial to subsequent accurate and efficient processing of overlapping cells.

[0073] Preferably, in a feasible implementation manner of this embodiment, for the method for obtaining the overlapping cell regions, please refer to Figure 3 , which shows a flowchart of a method for obtaining the overlapping cell regions provided in this embodiment. The method includes the following steps:

[0074] Step S201: Obtain the shape parameters of each cell region according to the shape of each cell region.

[0075] According to the shape parameter proposed by Zeng Qingnan in the paper "Research on Automatic Classification Technology of Red and White Blood Cell Images with Overlapping Cells Recognizable", the method for obtaining the shape parameter is as follows: for any cell region, the ratio of the area of the cell region to the square of the perimeter of the cell region is taken as the first value; the product of a specified constant and the first value is taken as the shape parameter of the cell region; where the specified constant is It should be noted that the research on automatic classification technology of red and white blood cell images with overlapping cells recognizable is well-known content and will not be elaborated here. Among them, the methods for obtaining the area and perimeter of the cell region are well-known technologies and will not be elaborated here.

[0076] Among them, the calculation formula for the shape parameter is: In the formula, is the shape parameter of the i-th cell region; is the area of the i-th cell region; is the perimeter of the i-th cell region.

[0077] It should be noted that the closer it is to 1, the closer the shape of the i-th cell region is to a circle, and the less likely the i-th cell region is an overlapping cell region; the closer it is to 0, the less likely the shape of the i-th cell region is a circle, and the more likely the i-th cell region is an overlapping cell region.

[0078] Thus, the shape parameter of each cell region is obtained.

[0079] Step S202: Obtain the edge undulation degree of each cell region according to the curvature fluctuation of the edge pixel points in each cell region.

[0080] When the fluctuation degree of the curvature of the edge pixel points in a certain cell region is greater, there is more likely a cell overlapping situation in the cell region. Therefore, in this embodiment, the edge undulation degree of each cell region is obtained according to the curvature fluctuation of the edge pixel points in each cell region. The greater the edge undulation degree, the more likely the corresponding cell region is an overlapping cell region.

[0081] Preferably, in an implementable manner of this embodiment, the method for obtaining the edge undulation degree is as follows: for any cell region, the curvature of each edge pixel point in the cell region is obtained as the reference curvature; the difference between the maximum reference curvature and the minimum reference curvature is taken as the first fluctuation degree; the greater the first fluctuation degree, the more irregular the edge in the cell region; in order to accurately characterize the undulation degree of the edge line in the cell region, the product of the variance of the reference curvature and the first fluctuation degree is taken as the edge undulation degree of the cell region.

[0082] Thus, the edge undulation degree of each cell region is obtained.

[0083] Step S203: Obtain the overlapping degree of each cell region according to the shape parameter and the degree of edge undulation of each cell region; wherein, the shape parameter and the overlapping degree are negatively correlated, and the degree of edge undulation and the overlapping degree are positively correlated.

[0084] It is known that the larger the shape parameter, the less likely the corresponding cell region is an overlapping cell region; the larger the degree of edge undulation, the more likely the corresponding cell region is an overlapping cell region. Furthermore, in this embodiment, the overlapping degree of each cell region is obtained according to the shape parameter and the degree of edge undulation of each cell region; wherein, the shape parameter and the overlapping degree are negatively correlated, and the degree of edge undulation and the overlapping degree are positively correlated. The larger the overlapping degree, the more likely the corresponding cell region is an overlapping cell region. Among them, the calculation formula of the overlapping degree is: ; in the formula, is the overlapping degree of the i-th cell region; is the shape parameter of the i-th cell region; is the degree of edge undulation of the i-th cell region; norm is a normalization function. In other embodiments, can be obtained through obtain , which is not limited here.

[0085] So far, the overlapping degree of each cell region is obtained.

[0086] Step S204: When the overlapping degree is greater than the preset overlapping degree threshold, regard the corresponding cell region as an overlapping cell region.

[0087] It is known that the larger the overlapping degree, the more likely the corresponding cell region is an overlapping cell region. Therefore, in this embodiment, the preset overlapping degree threshold is set to 0.5, and the implementer can set the size of the preset overlapping degree threshold according to the actual situation, which is not limited here. When the overlapping degree is greater than the preset overlapping degree threshold, regard the corresponding cell region as an overlapping cell region.

[0088] So far, the overlapping cell regions in the cell display image are screened out.

[0089] Step S3: Obtain the exposed edge lines of the covered cells in the overlapping cell region according to the length of each edge line in the overlapping cell region, the curvature of the edge pixel points on each edge line, and the curvature of the intersection pixel points of each edge line and other edge lines.

[0090] In the overlapping cell region, there is a cell that is partially covered by another cell. As Figure 4 shown is a local image of cell overlap, from Figure 4It can be seen that a part of the edge line of the covered cell, that is, the bare edge line, is normally displayed in the cell display image, and the length of the bare edge line is shorter than that of the complete edge line of the uncovered cell. At the same time, since the covered cell is covered by another cell on its surface, the degree of undulation of the edge line at the junction of the bare edge line of the covered cell and other edge lines is relatively large, that is, the curvature of the intersection pixel points of the bare edge line of the covered cell and other edge lines is significantly greater than the curvature of the edge pixel points on the bare edge line of the covered cell (excluding the intersection pixel points of the bare edge line of the covered cell and other edge lines). Therefore, in this embodiment, according to the length of each edge line in the overlapping cell region, the curvature of the edge pixel points on each edge line, and the curvature of the intersection pixel points of each edge line and other edge lines, the bare edge line of the covered cell in the overlapping cell region is obtained.

[0091] Preferably, in a realizable manner of this embodiment, for the method of obtaining the bare edge line, please refer to Figure 5 , which shows a flowchart of a method for obtaining a bare edge line provided in this embodiment. The method includes the following steps:

[0092] Step S301: Obtain the exposure degree of each edge line in each overlapping cell region according to the length of each edge line in each overlapping cell region, the curvature of the edge pixel points on each edge line, and the curvature of the intersection pixel points of each edge line and other edge lines.

[0093] When the length of a certain edge line in a certain overlapping cell region is short, and the curvature of the intersection pixel points of this edge line and other edge lines is significantly greater than the curvature of the edge pixel points on this edge line, it indicates that this edge line is more likely to be the bare edge line of the covered cell. It should be noted that the edge pixel points on this edge line do not include the intersection pixel points of this edge line and other edge lines. Therefore, in this embodiment, according to the length of each edge line in each overlapping cell region, the curvature of the edge pixel points on each edge line, and the curvature of the intersection pixel points of each edge line and other edge lines, the exposure degree of each edge line in each overlapping cell region is obtained. The greater the exposure degree, the more likely the corresponding edge line is the bare edge line of the covered cell.

[0094] Preferably, in an implementable manner of this embodiment, the method for obtaining the exposure degree is as follows: for any edge line in any overlapping cell region, obtain the average value of the curvatures of all edge pixels on the edge line as the curvature analysis value of the edge line; obtain the average value of the curvatures of the intersection pixels of the edge line and other edge lines in the overlapping cell region as the curvature reference value; the greater the difference between the curvature analysis value and the curvature reference value, the more likely the edge line is an exposed edge line. Furthermore, obtain the absolute value of the difference between the curvature analysis value and the curvature reference value as the first difference; the greater the first difference, the more likely the edge line is an exposed edge line. At the same time, when the edge line is shorter, that is, the number of edge pixels on the edge line is smaller, it also indicates that the edge line is more likely to be an exposed edge line. Furthermore, in this embodiment, the result of normalizing the ratio of the first difference to the number of edge pixels on the edge line is used as the exposure degree of the edge line. Among them, the calculation formula for the exposure degree is: ; where is the exposure degree of the b-th edge line in the a-th overlapping cell region; is the curvature analysis value of the b-th edge line in the a-th overlapping cell region; is the curvature reference value corresponding to the b-th edge line in the a-th overlapping cell region; is the number of edge pixels on the b-th edge line in the a-th overlapping cell region; norm is a normalization function; is the first difference; is the absolute value function.

[0095] So far, obtain the exposure degree of each edge line in each overlapping cell region.

[0096] Step S302: When the exposure degree is greater than the preset exposure degree threshold, regard the corresponding edge line as an exposed edge line.

[0097] It is known that the greater the exposure degree, the more likely the corresponding edge line is the exposed edge line of the covered cell. Therefore, in this embodiment, the preset exposure degree threshold is set to 0.6. The implementer can set the size of the preset exposure degree threshold according to the actual situation, which is not limited here. When the exposure degree is greater than the preset exposure degree threshold, regard the corresponding edge line as an exposed edge line.

[0098] So far, determine the exposed edge lines of the covered cells in each overlapping cell region.

[0099] In another embodiment, considering that due to the instability of environmental lighting conditions, noise may exist in the cell display image. Noise usually appears as irregular bright spots or dark patches, and there may also be corresponding edge lines. Therefore, the exposed edge lines may be pseudo-edge lines formed by noise. In this embodiment, in order to accurately classify the cells in the cell display image, it is necessary to determine the morphology of each cell in the cell display image. Therefore, it is necessary to further analyze the exposed edge lines to avoid the pseudo-edge lines formed by noise from affecting the accuracy of the subsequent automatic classification and detection of peripheral blood cell morphology. It is known that noise is randomly distributed bright spots or dark patches, and the brightness change between the area corresponding to the noise and its surrounding background area is abrupt. Therefore, there is an obvious difference between the gray value of the edge pixel points on the pseudo-edge line corresponding to the noise and the gray values of other pixel points in its local background area. Because the cell surface usually has a certain transparency or film structure, the gray value difference between the edge pixel points on the cell edge line and the pixel points in its local background area is smoothly changing. Therefore, in this embodiment, the real exposed edge line is obtained according to the gray value difference between each edge pixel point on each exposed edge line and its preset neighborhood pixel points.

[0100] Preferably, in a feasible implementation manner of this embodiment, the method for obtaining the real exposed edge line is as follows: for any edge pixel point on any exposed edge line, the straight line perpendicular to the tangent line of the edge pixel point is used as the target straight line, and the preset number of pixel points located on the target straight line and closest to the edge pixel point outside the covered cell corresponding to the exposed edge line are used as the preset neighborhood pixel points of the edge pixel point. In this embodiment, the preset number is set to 20, and the implementer can set the size of the preset number according to the actual situation, which is not limited here. The absolute value of the difference between the gray value of the edge pixel point and the average gray value of its preset neighborhood pixel points is obtained as the brightness mutation degree of the edge pixel point; the greater the brightness mutation degree, the more likely the edge pixel point is generated by noise. In order to comprehensively analyze whether the exposed edge line is a real exposed edge line, the negative correlation and normalization result of the sum of the brightness mutation degrees of all edge pixel points on the exposed edge line is used as the brightness smoothness degree of the exposed edge line; the greater the brightness smoothness degree, the more likely the exposed edge line is the real exposed edge line corresponding to the cell. Therefore, in this embodiment, the preset brightness smoothness degree threshold is set to 0.7, and the implementer can set the size of the preset brightness smoothness degree threshold according to the actual situation, which is not limited here. When the brightness smoothness degree is greater than the preset brightness smoothness degree threshold, the corresponding exposed edge line is used as the real exposed edge line. Thus, the real exposed edge line in the overlapping cell region is determined, effectively avoiding the interference of noise.

[0101] Among them, the calculation formula for the brightness smoothness degree is: ; in the formula, is the brightness smoothness of the m-th bare edge line in the v-th overlapping cell region; is the number of upper edge pixels on the m-th bare edge line in the v-th overlapping cell region; is the gray value of the r-th edge pixel on the m-th bare edge line in the v-th overlapping cell region; is the average gray value of the preset neighborhood pixels of the r-th edge pixel on the m-th bare edge line in the v-th overlapping cell region; is the brightness mutation degree of the r-th edge pixel on the m-th bare edge line in the v-th overlapping cell region; is the absolute value function; exp is the exponential function with the natural constant as the base.

[0102] It should be noted that in this embodiment, the subsequent bare edge lines are all defaulted to real bare edge lines, effectively avoiding the influence of noise generation.

[0103] Step S4: According to the curvature and endpoints of each bare edge line, obtain the virtual coverage edge line of the covered cell corresponding to each bare edge line, determine the complete cell edge line of each covered cell, and perform automatic classification detection of peripheral blood cell morphology.

[0104] It is known that the automatic classification detection of peripheral blood cell morphology needs to be classified according to the complete morphology of each cell. Therefore, in this embodiment, it is necessary to obtain the edge line covered by the covered cell corresponding to each bare edge line. The edge line covered by the covered cell should be consistent with the curvature change trend of the bare edge line of the covered cell. Therefore, in this embodiment, according to the curvature and endpoints of each bare edge line, the virtual coverage edge line of the covered cell corresponding to each bare edge line, that is, the edge line covered by the covered cell, is obtained. Among them, the method of obtaining the curvature of the edge line is a well-known technology and will not be elaborated.

[0105] Preferably, in an implementable manner of this embodiment, the method for obtaining the virtual coverage edge line is as follows: for any exposed edge line, the pixel points at both ends of the exposed edge line are used as target pixel points; for any target pixel point, the eight-neighbor pixel points of the target pixel point that the exposed edge line does not pass through are used as the adjacent pixel points of the target pixel point, ensuring that the adjacent pixel points of the target pixel point are definitely not passed by the exposed edge line; further, the exposed edge line is smoothly connected to each adjacent pixel point of the target pixel point through the Bezier curve technology to obtain the reference edge line corresponding to each adjacent pixel point of the target pixel point; among them, the Bezier curve technology is a well-known technology and will not be elaborated here. The absolute value of the difference in curvature between the reference edge line corresponding to each adjacent pixel point of the target pixel point and the exposed edge line is used as the second difference; the smaller the second difference, the more consistent the curvature change trend of the corresponding reference edge line and the exposed edge line. Therefore, in this embodiment, the adjacent pixel points on the reference edge line corresponding to the smallest second difference are used as the first virtual edge pixel points of the exposed edge line; it should be noted that there must be only one smallest second difference;

[0106] The exposed edge line is smoothly connected to the first virtual edge pixel point through the Bezier curve technology to obtain the first reference exposed edge line; the eight-neighbor pixel points of the first virtual edge pixel point that the first reference exposed edge line does not pass through are used as the adjacent pixel points of the first virtual edge pixel point; the first reference exposed edge line is smoothly connected to each adjacent pixel point of the first virtual edge pixel point through the Bezier curve technology to obtain the reference edge line corresponding to each adjacent pixel point of the first virtual edge pixel point; the absolute value of the difference in curvature between the reference edge line corresponding to each adjacent pixel point of the first virtual edge pixel point and the first reference exposed edge line is used as the third difference; the smaller the third difference, the more consistent the curvature change trend of the corresponding reference edge line and the first reference exposed edge line. Therefore, the adjacent pixel points on the reference edge line corresponding to the smallest third difference are used as the second virtual edge pixel points of the exposed edge line; it should be noted that there must be only one smallest third difference; and so on, each virtual edge pixel point of the exposed edge line is obtained until the other target pixel point of the exposed edge line is reached; finally, the virtual edge pixel points obtained with each target pixel point as the starting point are jointly curve-fitted by the least squares method, and the obtained curve is used as the virtual coverage edge line of the covered cell corresponding to the exposed edge line. Among them, the least squares method is a well-known technology and will not be elaborated here.

[0107] So far, the virtual coverage edge line of the covered cell corresponding to each exposed edge line is obtained.

[0108] The bare edge line of each covered cell and the virtual covered edge line are combined into a closed edge line to determine the complete cell edge line of each covered cell, that is, to determine the morphology of each covered cell, which is beneficial to accurately identify and classify the cells in the cell display image subsequently, and effectively solves the problem of inaccurate cell classification caused by cell overlap. Finally, using a convolutional neural network deep learning model, according to the morphology of each cell in the cell display image, the cells in the cell display image are automatically identified and classified, and the full-automatic classification detection of peripheral blood cell morphology is accurately completed. Among them, the convolutional neural network deep learning model is a well-known technology and will not be elaborated here.

[0109] In summary, in this embodiment, the cell regions in the cell display image of peripheral blood are obtained; according to the morphology of the cell regions and the curvature of the edge pixel points in the cell regions, the overlapping cell regions are obtained; according to the length of each edge line in the overlapping cell regions, the curvature of the edge pixel points on each edge line, and the curvature of the intersection pixel points of each edge line and other edge lines, the bare edge lines of the covered cells are obtained; according to the curvature and endpoints of the bare edge lines, the complete cell edge lines of the covered cells are determined, and the full-automatic classification detection of peripheral blood cell morphology is carried out. By obtaining the complete cell edge lines of the covered cells, the present invention determines the morphology of the covered cells, avoids the inaccurate situation of automatic cell recognition and classification caused by cell overlap, and ensures the accuracy of the full-automatic classification detection of peripheral blood cell morphology.

[0110] Embodiment 2:

[0111] The present invention also proposes a full-automatic classification detection system for peripheral blood cell morphology. Please refer to Figure 6 , which shows the structure diagram of a full-automatic classification detection system for peripheral blood cell morphology provided by an embodiment of the present invention. The system includes: an acquisition module 10, an overlapping cell region acquisition module 20, a bare edge line acquisition module 30, and a detection module 40.

[0112] The acquisition module 10 is used to acquire the cell regions in the cell display image of peripheral blood.

[0113] The overlapping cell region acquisition module 20 is used to acquire the overlapping cell regions according to the morphology of each cell region and the curvature fluctuation of the edge pixel points in each cell region.

[0114] The bare edge line acquisition module 30 is used to acquire the bare edge lines of the covered cells in the overlapping cell regions according to the length of each edge line in the overlapping cell regions, the curvature of the edge pixel points on each edge line, and the curvature of the intersection pixel points of each edge line and other edge lines.

[0115] The detection module 40 is configured to obtain the virtual covered edge lines of the covered cells corresponding to each exposed edge line according to the curvature and endpoints of each exposed edge line, determine the complete cell edge lines of each covered cell, and perform automatic classification detection of peripheral blood cell morphology.

[0116] It should be noted that: for the system provided in the above embodiment, only the division of the above functional modules is used for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, an automatic classification detection system for peripheral blood cell morphology and an embodiment of an automatic classification detection method for peripheral blood cell morphology provided in the above embodiment belong to the same concept. The specific implementation process can be seen in the method embodiment and will not be elaborated here.

[0117] Embodiment 3:

[0118] The present invention also provides an automatic classification detection device for peripheral blood cell morphology. The device includes a memory and a processor. Among them, executable program code is stored in the memory, and the processor is configured to call and execute the executable program code to execute an automatic classification detection method for peripheral blood cell morphology provided in an embodiment of the present application. The device may specifically be a chip, a component or a module. The chip may include a processor and a memory connected thereto; among them, the memory is used to store instructions. When the processor calls and executes the instructions, the chip can execute an automatic classification detection method for peripheral blood cell morphology provided in the above embodiment.

[0119] In addition, an embodiment of the present application also protects a computer device. Please refer to Figure 7 , the computer device includes a memory 401, a processor 402, and a computer program 403 stored in the memory 401 and running on the processor 402. Among them, when the processor 402 executes the computer program 403, the computer device can execute any one of the automatic classification detection methods for peripheral blood cell morphology introduced above.

[0120] Embodiment 4:

[0121] This embodiment also provides a computer-readable storage medium. Computer program code is stored in the computer-readable storage medium. When the computer program code runs on a computer, the computer is enabled to execute the above-related method steps to implement an automatic classification detection method for peripheral blood cell morphology provided in the above embodiment.

[0122] Embodiment 5:

[0123] This embodiment also provides a computer program product. When the computer program product runs on a computer, it causes the computer to execute the above-related steps to implement a fully automatic classification and detection method for peripheral blood cell morphology provided by the above embodiment.

[0124] Among them, the device, computer-readable storage medium, computer program product, or chip provided in this embodiment are all used to execute the corresponding method provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding method provided above, and will not be elaborated here.

[0125] It should be noted that the above sequence of embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the 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.

[0126] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. An automatic classification detection method for peripheral blood cell morphology, characterized in that, The method includes the following steps: Obtain the cell regions in the cell display image of peripheral blood; Obtain the overlapping cell regions according to the morphology of each cell region and the curvature fluctuation of the edge pixel points in each cell region; Obtain the exposed edge lines of the covered cells in the overlapping cell regions according to the length of each edge line in the overlapping cell regions, the curvature of the edge pixel points on each edge line in the overlapping cell regions, and the curvature of the intersection pixel points of each edge line in the overlapping cell regions with other edge lines; Obtain the virtual covered edge lines corresponding to each exposed edge line according to the curvature and endpoints of each exposed edge line, determine the complete cell edge lines of each covered cell, and perform automatic classification detection of the morphology of peripheral blood cells; The method for obtaining the overlapping cell regions is as follows: Obtain the shape parameters of each cell region according to the shape of each cell region; the shape parameters are used to describe the degree to which the cell morphology is circular; Obtain the edge undulation degree of each cell region according to the curvature fluctuation of the edge pixel points in each cell region; Obtain the overlapping degree of each cell region according to the shape parameters and edge undulation degree of each cell region; among them, the shape parameters and the overlapping degree are negatively correlated, and the edge undulation degree and the overlapping degree are positively correlated; When the overlapping degree is greater than the preset overlapping degree threshold, the corresponding cell region is used as the overlapping cell region; The method for obtaining the exposed edge lines is as follows: Obtain the exposure degree of each edge line in each overlapping cell region according to the length of each edge line in each overlapping cell region, the curvature of the edge pixel points on each edge line, and the curvature of the intersection pixel points of each edge line with other edge lines in the overlapping cell regions; When the exposure degree is greater than the preset exposure degree threshold, the corresponding edge line is used as the exposed edge line; The method for obtaining the exposure degree is as follows: For any edge line in any overlapping cell region, obtain the average value of the curvatures of all edge pixel points on this edge line in the overlapping cell region, and use it as the curvature analysis value of this edge line in the overlapping cell region; Obtain the average value of the curvatures of the intersection pixel points of this edge line with other edge lines in the overlapping cell region, and use it as the curvature reference value of this edge line in the overlapping cell region; Take the difference between the curvature analysis value and the curvature reference value as the first difference; Take the result of normalizing the ratio of the first difference to the number of edge pixel points on this edge line as the exposure degree of this edge line.

2. The full-automatic classification detection method for peripheral blood cell morphology according to claim 1, characterized in that The method for obtaining the shape parameters is as follows: For any cell region, take the ratio of the area of this cell region to the square of the perimeter of this cell region as the first value; Take the product of the specified constant and the first value as the shape parameter of this cell region.

3. The full-automatic classification detection method for peripheral blood cell morphology according to claim 1, characterized in that, The method for obtaining the edge undulation degree is as follows: For any cell region, obtain the curvature of each edge pixel point in this cell region, and use them all as reference curvatures; Take the difference between the maximum reference curvature and the minimum reference curvature as the first fluctuation degree; Take the product of the variance of the reference curvatures and the first fluctuation degree as the edge undulation degree of this cell region.

4. The full-automatic classification detection method for the morphology of peripheral blood cells according to claim 1, wherein The method for obtaining the exposed edge lines further includes: Obtain the true exposed edge line according to the gray - level difference between each edge pixel on each exposed edge line and its preset neighborhood pixels.

5. The full-automatic classification detection method for peripheral blood cell morphology according to claim 4, characterized in that, The method for obtaining the true exposed edge line is as follows: For any edge pixel on any exposed edge line, take the line perpendicular to the tangent of this edge pixel as the target line, and take the preset number of pixels closest to this edge pixel on the target line and outside the covered cell corresponding to this exposed edge line as the preset neighborhood pixels of this edge pixel; Obtain the difference between the gray - level value of this edge pixel and the average gray - level value of its preset neighborhood pixels as the degree of brightness mutation of this edge pixel; Take the result of negative correlation and normalization of the sum of the degrees of brightness mutation of all edge pixels on this exposed edge line as the brightness smoothness of this exposed edge line; When the brightness smoothness is greater than the preset brightness smoothness threshold, take the corresponding exposed edge line as the true exposed edge line.

6. The full-automatic classification detection method for peripheral blood cell morphology according to claim 1, wherein, The method for obtaining the virtual covered edge line is as follows: For any exposed edge line, take the pixels at both ends of this exposed edge line as target pixels; For any target pixel, take the eight - neighborhood pixels of this target pixel that this exposed edge line does not pass through as the adjacent pixels of this target pixel; Smoothly connect this exposed edge line with each adjacent pixel of this target pixel through Bezier curve technology to obtain the reference edge line corresponding to each adjacent pixel of this target pixel; Obtain the difference in curvature between the reference edge line corresponding to each adjacent pixel of this target pixel and this exposed edge line as the second difference; Take the adjacent pixel on the reference edge line corresponding to the minimum second difference as the first virtual edge pixel of this exposed edge line; Smoothly connect this exposed edge line with the first virtual edge pixel through Bezier curve technology to obtain the first reference exposed edge line; Take the eight - neighborhood pixels of the first virtual edge pixel that the first reference exposed edge line does not pass through as the adjacent pixels of the first virtual edge pixel; Smoothly connect the first reference exposed edge line with each adjacent pixel of the first virtual edge pixel through Bezier curve technology to obtain the reference edge line corresponding to each adjacent pixel of the first virtual edge pixel; Obtain the difference in curvature between the reference edge line corresponding to each adjacent pixel of the first virtual edge pixel and the first reference exposed edge line as the third difference; Take the adjacent pixel on the reference edge line corresponding to the minimum third difference as the second virtual edge pixel of this exposed edge line; And so on, obtain each virtual edge pixel of this exposed edge line until reaching the other target pixel of this exposed edge line; Curve - fit the virtual edge pixels obtained with each target pixel as the starting point by the least - squares method, and take the obtained curve as the virtual covered edge line of the covered cell corresponding to this exposed edge line.

7. The full-automatic classification detection method for the morphology of peripheral blood cells according to claim 1, wherein, The method for obtaining the cell region is as follows: Obtain the cell regions in the cell display image through the Canny edge detection algorithm and the connected component algorithm.

Images