Method for detecting, identifying and counting cerebrospinal fluid cell images

Abstract

The invention discloses a method for detecting, identifying and counting cerebrospinal fluid cell images, which comprises the following steps of: firstly, acquiring a high-quality digital image of a cerebrospinal fluid cell smear through a microscope, and then revealing edges and other areas for distinguishing cells in the image through gradient value calculation; furthermore, through calculation of a local contrast change rate, microscopic change of the cell image is accurately captured, and normal cells and cells with abnormal forms are effectively distinguished; and furthermore, normal cells and abnormal cells are counted to obtain a counting result, so that medical personnel can be better helped to identify the illness state of the patient, and powerful patient medical record information is provided for subsequent medical assistance.

Description

Technical Field

[0001] The invention relates to the field of cell counting, and in particular to a method for detecting, identifying and counting cerebrospinal fluid cell images. Background Art

[0002] Cerebrospinal fluid cytology examination is the "gold standard" for the diagnosis of meningeal carcinomatosis and is of great significance for the early diagnosis of meningeal carcinomatosis.

[0003] Cytology report To issue a reliable cytology diagnosis report, it is necessary to understand the patient's clinical manifestations, diagnosis and treatment process and related laboratory test results. After comprehensive analysis, the cytology manifestations should be described and laboratory diagnosis opinions or suggestions should be given.

[0004] The laboratory test results are mainly based on the image detection and identification and counting of cerebrospinal fluid cells to obtain the laboratory test results.

[0005] Nowadays, the image detection of cerebrospinal fluid cells is mainly based on the image detection and identification of cerebrospinal fluid cells by machine learning algorithms, and the efficiency and accuracy of cell counting are improved by automated image processing and analysis technology. At present, the cerebrospinal fluid cell image recognition method based on machine learning generally adopts deep learning algorithms such as convolutional neural networks (CNN) to detect and classify cells through training models. These methods can improve the speed of cell recognition to a certain extent, but still face some technical defects, especially when the cerebrospinal fluid cells in the cerebrospinal fluid cell image mutate and produce convex or concave conditions, the machine learning algorithm cannot accurately identify the edges of such cerebrospinal fluid cell images, and thus cannot accurately identify the complete cerebrospinal fluid cell image, so it is impossible to accurately count cerebrospinal fluid cells, which in turn causes great difficulties in obtaining cytological diagnosis reports. Summary of the invention

[0006] The purpose of the present invention is to provide a method for detecting, identifying and counting cerebrospinal fluid cell images, which solves the above-mentioned technical problems pointed out in the prior art.

[0007] The present invention provides a method for detecting, identifying and counting cerebrospinal fluid cell images, comprising the following steps: Acquire digital images of cerebrospinal fluid cell smears; Obtaining a digital image pixel gradient value corresponding to each digital image pixel according to a grayscale value difference between adjacent digital image pixels of each digital image pixel in the digital image of the cerebrospinal fluid cell smear; Analyze the local directional gradient in the preset neighborhood window of each digital image pixel to obtain the local contrast change rate of each digital image pixel; Based on the local contrast change rate combined with the degree of abnormal cell morphology and the digital image pixel gradient value, the cell image in the digital image is identified and counted to obtain a counting result; The counting result includes the number of normal cell types counted and the number of abnormal cell types counted.

[0008] Preferably, the identifying and counting of the cell images in the digital image based on the local contrast change rate combined with the degree of abnormal cell morphology and the digital image pixel gradient value to obtain the counting result comprises the following steps: Acquire a connection area of ​​digital image pixel points according to the local contrast change rate, and obtain the degree of abnormal cell morphology of the cerebrospinal fluid cell smear according to the distribution characteristics corresponding to the connection area; According to the degree of abnormal cell morphology and the local contrast change rate corresponding to each digital image pixel, the value of the influence factor of the region on the pixel change is obtained by analysis and calculation; Obtaining a cell characteristic value of each pixel point of the digital image according to the digital image pixel point gradient value and the value of the influence factor of the region on the pixel point change; Clustering is performed based on the cell feature values ​​to obtain multiple clusters; based on the clusters and the connection areas, the clusters and the connection areas are merged and supplemented in combination with a region growing algorithm to obtain multiple cell images; and based on the cell images and the degree of abnormal cell morphology, the cell images are counted to obtain counting results.

[0009] Preferably, the calculation method of the value of the influence factor of the region on the pixel change is: ; In the formula, and is the weight coefficient; The degree of abnormal cell morphology; is the local contrast change rate; The cell characteristic value is calculated as follows: ; In the formula, is the gradient value of the pixel point of the digital image; is the value of the influence factor of the region on the pixel change; and is the weight coefficient, and .

[0010] Preferably, the step of merging and supplementing the clusters and the connected regions based on the clusters and the connected regions in combination with a region growing algorithm to obtain a plurality of cell images comprises the following steps: Acquire a first cell image based on each of the clusters; Performing region growing on edge pixel points corresponding to the first cell image to obtain a second cell image; Constructing a boundary point set based on second edge pixel points of the second cell image; Calculate and obtain the spatial fusion degree R of the boundary point set and the connection area; The second cell image and the connection area are merged based on the spatial fusion degree R to obtain a cell image.

[0011] Preferably, performing region growing on edge pixels corresponding to the first cell image to obtain the second cell image comprises the following steps: Based on the first cell image and the digital image, the first evaluation value and the second evaluation value are obtained by analyzing and calculating after labeling; An iterative analysis region growing operation is performed based on the first judgment value and the second judgment value in combination with the edge pixel points to obtain a second cell image.

[0012] Preferably, the step of analyzing and calculating the first evaluation value and the second evaluation value after labeling the first cell image and the digital image comprises the following steps: Labeling each of the first cell images to obtain a cell image label; Obtaining cell image labels of first cell pixels corresponding to each of the first cell images based on the cell image labels; Based on all digital pixels in the digital image, a plurality of unlabeled pixels are obtained by removing all the first cell pixels; Constructing an unlabeled pixel point set based on all the unlabeled pixel points; The first influencing evaluation factor is obtained by calculating based on the brightness value of each of the first cell pixels in the first cell image. ; Calculate the second influencing factor based on the brightness value of the unlabeled pixel ; The first influencing evaluation factor is based on the brightness value of each of the first cell pixels in the first cell image. Calculate the first judgment value Based on the brightness value of the unlabeled pixel and the second influencing factor Calculate the second judgment value .

[0013] Preferably, the first influencing factor The calculation method is: ; Where y is the brightness value of the first cell pixel; is the total number of pixels; The total number of pixels in the first cell; The second influencing factor The calculation method is: ; Where x is the brightness value of the unlabeled pixel; e is the total number of unlabeled pixels; The calculation method of the first evaluation value is: ; The second evaluation value is calculated as follows: .

[0014] Preferably, the iterative analysis of the region growing operation based on the first evaluation value and the second evaluation value combined with the edge pixel points to obtain the second cell image comprises the following steps: Based on the coordinates of each digital image pixel point in the digital image and the brightness value of each digital image pixel point, a multi-factor analysis is performed in combination with the edge pixel points to obtain an edge pixel point sequence set; based on the order of each edge pixel point in the edge pixel point sequence set, each edge pixel point is sequentially traversed, and a first local brightness contrast is calculated based on each unlabeled pixel point in the 8 neighboring pixel points corresponding to the ti-th edge pixel point; Calculate the second local brightness contrast between the neighboring pixel point corresponding to the cth unlabeled pixel point in the current neighboring pixel points and the tith edge pixel point; and select a number of the unlabeled pixel points less than or equal to the first local brightness contrast as candidate pixel points based on the second local brightness contrast; Determine whether the number a of the candidate pixel points is greater than or equal to a preset minimum number threshold of the candidate pixel points; if so, use the candidate pixel points as supplementary pixel points to be determined; Obtain a new first cell image to be determined based on the supplementary pixel points to be determined and the first cell image; obtain a new unlabeled pixel point set to be determined based on the new first cell image pixel points to be determined corresponding to the new first cell image to be determined and the digital image; A first judgment value change rate is calculated based on the brightness value of each pixel point of the new first cell image to be determined in the new first cell image to be determined and the first judgment value; a second judgment value change rate is calculated based on the brightness value of each new unlabeled pixel point in the new unlabeled pixel point set to be determined and the second judgment value; Determine whether the change rate of the first judgment value is less than the change rate of the second judgment value. If so, determine the to-be-determined supplementary pixel point as the target supplementary pixel point, and merge the target supplementary pixel point into the first cell image to obtain a second cell image; and obtain a new unlabeled pixel point set based on the second cell image; update the second judgment value based on the brightness value of each second cell image pixel point in the second cell image or the brightness value of each new unlabeled pixel point in the new unlabeled pixel point set to obtain a new second judgment value; Determine whether all edge pixel points have been traversed; if so, stop the iteration and output each of the second cell images; if not, return to the above step to continue the traversal operation until all edge pixel points have been traversed and multiple second cell images are output.

[0015] Preferably, the step of performing a multi-factor analysis based on the coordinates of each digital image pixel point in the digital image and the brightness value of each digital image pixel point in combination with the edge pixel points to obtain an edge pixel point sequence set comprises the following steps: Traversing each of the edge pixels, and calculating the edge pixel contrast coefficient F based on the brightness value of the edge pixel and the neighborhood brightness values ​​of the neighborhood pixels of the edge pixel; Calculating a regional connectivity feature C based on the number of first cell pixels in the neighborhood pixels of the edge pixel, the number of window pixels within a preset window range of the edge pixel, and the number of first cell pixels in the window pixel; Obtain the coordinates of each of the first cell pixel points, and calculate the center point coordinates of each of the first cell images and the average radius R of the first cell image based on the coordinates of the first cell pixel points; calculate the center distance based on the coordinates of the edge pixel points and the center point coordinates of the first cell image corresponding to the edge pixel points; calculate the regional pixel density based on the number of first cell pixel points within the preset window range; calculate the distribution feature D based on the regional pixel density, the average radius R, the center distance and the preset target density; The edge pixel processing priority L is calculated based on the edge pixel contrast coefficient F, the region connectivity feature C and the distribution feature D; The edge pixel points are sorted from high to low based on the edge pixel point processing priority L to obtain an edge pixel point sequence set.

[0016] Preferably, the edge pixel contrast coefficient is calculated as follows: ; In the formula, is the brightness value of the edge pixel; is the average brightness value of the neighboring pixels of the edge pixel; G is the gradient value of the edge pixel; is the maximum gradient value in the digital image; The calculation method of the regional connectivity feature C is: ; In the formula, is the number of first cell pixels in the neighborhood of edge pixels; is the number of the first cell pixel in the window pixel; is the number of pixels in the window; The distribution characteristic D is calculated as follows: ; Where d is the center distance; is the average radius; is the regional pixel density; is the target density; The edge pixel processing priority L is calculated as follows: ; Where α1, β1 and γ1 are weight coefficients.

[0017] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages: By analyzing the above-mentioned cerebrospinal fluid cell image detection, identification and counting method provided by the present invention, it can be known that in specific application, a high-quality digital image of a cerebrospinal fluid cell smear is first obtained through a microscope, and then the edges and other areas of the image for distinguishing cells are revealed by calculating the gradient value; further, by calculating the local contrast change rate, the microscopic changes of the cell image are accurately captured, and normal cells and morphologically abnormal cells are effectively distinguished; then, the normal cells and abnormal cells are counted to obtain the counting results, which can better help medical staff identify the patient's condition and provide powerful patient medical record information for subsequent medical assistance. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of the main process of a method for detecting, identifying and counting cerebrospinal fluid cell images; Figure 2 A schematic diagram of the simulation of the morphology of intracranial medulloblastoma cells in a method for detecting, identifying and counting cerebrospinal fluid cell images; Figure 3 A schematic diagram of an operation flow of obtaining a second cell image in a method for detecting, identifying and counting cerebrospinal fluid cell images; Figure 4 A schematic diagram of a simulation of the edge of an unrecognized cell image in a method for detecting, identifying and counting cerebrospinal fluid cell images. DETAILED DESCRIPTION

[0019] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] The present invention is further described in detail below through specific embodiments in conjunction with the accompanying drawings.

[0021] like Figure 1 As shown, an embodiment of the present invention provides a method for detecting, identifying and counting cerebrospinal fluid cell images, comprising the following steps: Step S10: obtaining a digital image of a cerebrospinal fluid cell smear; It should be noted that the above-mentioned embodiment of the present application first obtains a high-resolution digital image of the cell smear through a microscope (the microscope can be an electronic cytometer or an electron microscope with electronic high-definition image photography and recording and video recording functions). Through the high-quality image, the morphology, structure and other characteristics of the cell can be captured, providing clear basic data for subsequent analysis.

[0022] Step S20: obtaining a digital image pixel gradient value corresponding to each digital image pixel according to a grayscale value difference between adjacent digital image pixels of each digital image pixel in the digital image of the cerebrospinal fluid cell smear; It should be noted that in the above-mentioned embodiments of the present application, the gradient value is an important indicator for measuring the change in pixel brightness in an image. Generally, the edge or boundary area in the image has a large gray value change, and a large gradient value can help distinguish the edge of the cell from other areas; Specifically, the embodiment of the present application quantifies the brightness change of each pixel by calculating the grayscale difference between adjacent pixels to obtain a gradient value. Places with larger gradient values ​​usually indicate cell edges or areas with more obvious structural changes.

[0023] Step S30: Analyze the local directional gradient in the preset neighborhood window of each digital image pixel to obtain the local contrast change rate of each digital image pixel; It should be noted that the above-mentioned embodiment of the present application further analyzes the local contrast change rate of each pixel in the image. By presetting a neighborhood window around each pixel and calculating the local directional gradient within the window, the local contrast changes in different areas of the cell image can be more accurately captured. This is a sign of an important feature of the analysis of cell morphological changes. The local contrast change rate reflects the texture changes in the local area of ​​the image and can be used to identify structural changes within the cell (i.e., the above-mentioned local contrast change rate, such as sharpening of cell boundaries, structural distortion, etc.); this information is very necessary for cell recognition and abnormal cell detection.

[0024] Step S40: identifying and counting the cell images in the digital image based on the local contrast change rate combined with the degree of abnormal cell morphology and the digital image pixel gradient value to obtain a counting result; The counting results include the number of normal cell types counted and the number of abnormal cell types counted; It should be noted that the above-mentioned embodiment of the present application recognizes and counts cells by analyzing the local contrast change rate in the image in combination with the gradient value in the image, recognizes the cell image from the image, and distinguishes it into normal cells and abnormal cells (abnormal cells refer to tumor cells, such as intracranial germ cell tumor cells, intracranial medulloblastoma cells, metastatic rhabdomyosarcoma cells, atypical teratoid rhabdoid tumor cells, etc.); The counting of cell images is based on the features of the image (such as morphological changes, boundary features, contrast changes, etc.). Through the combination of these features, normal cells and abnormal cells can be effectively detected and counted. Abnormal cells usually have irregular morphology, deformation or different local contrast change rates. These characteristics and features will be used to determine whether the cells are abnormal; Specifically, Figure 2 As shown, taking the morphology of intracranial medulloblastoma cells among abnormal cells as an example, it is manifested as tumor-like protrusions visible on part of the cell membrane. Conventional image segmentation processing operations are difficult to identify and segment these tumor-like protrusions. The technical solution adopted in the embodiment of the present application performs comprehensive recognition based on the morphological changes, boundary characteristics, and contrast changes of intracranial medulloblastoma cells, and more accurately identifies their edges, thereby accurately segmenting each intracranial medulloblastoma cell, thereby accurately and conveniently performing subsequent cell counting processing.

[0025] The above-mentioned embodiment of the present application first obtains a high-quality digital image of a cerebrospinal fluid cell smear through a microscope, and then reveals the edges and other areas of the image that distinguish cells by calculating the gradient value; further, by calculating the local contrast change rate, the microscopic changes of the cell image are accurately captured, and normal cells and morphologically abnormal cells are effectively distinguished; then, the normal cells and abnormal cells are counted to obtain the counting results, which can better help medical staff identify the patient's condition and provide powerful patient medical record information for subsequent medical assistance.

[0026] Specifically, in step S40, the cell image in the digital image is identified and counted based on the local contrast change rate combined with the degree of cell morphology abnormality and the digital image pixel gradient value to obtain a counting result, including the following steps: Step S41: acquiring a connection area of ​​digital image pixel points according to the local contrast change rate, and obtaining the degree of abnormal cell morphology of the cerebrospinal fluid cell smear according to the distribution characteristics corresponding to the connection area; It should be noted that the above-mentioned connected region refers to a continuous curve obtained by fitting the cosine similarity between the local contrast change rates of the above-mentioned adjacent digital image pixels in the image, and then, a connected region obtained according to the continuous curve; each "connected region" represents a cell or a part of a cell; The above distribution characteristics refer to the geometric distribution of these connection areas in the digital image, including information such as cell shape, size, and position. Normal cells may present a symmetrical and regular shape, while abnormal cells may show irregular shapes, size changes, deformations, and other characteristics; The degree of abnormal cell morphology: The degree of abnormal cell morphology is calculated based on the distribution characteristics of these connection areas; if the cell morphology undergoes pathological changes, such as cell deformation, different sizes, abnormal structures, etc., it means that the degree of abnormal cell morphology is high; the degree of morphological abnormality is quantified by comparing the morphological characteristics of normal cells and abnormal cells.

[0027] Step S42: Analyze and calculate the influence factor value of the region on the pixel change according to the abnormal degree of cell morphology and the local contrast change rate corresponding to each digital image pixel (i.e., set an influence factor value that can reflect the overall comprehensive change of the morphology of each cell pixel in the region); It should be noted that the above-mentioned embodiment of the present application calculates the value of the influence factor of the region on the change of the pixel point based on the local area where the pixel point is located (the local area is the local area corresponding to the degree of abnormal cell morphology of the above-mentioned connected area and the local contrast change rate of the local area), and the pixels in each area jointly affect the cell morphology change in the area and reflect the degree of abnormal cell morphology as a whole; in layman's terms, the connected area obtained by the continuous curve fitted by the local contrast change rate and the cosine similarity can be understood as a local area of ​​the cell, and the pixel points in this area can be comprehensively calculated according to the local characteristics. The value of the influence factor of the region on the change of the pixel point; The calculation method of the influence factor value of the above area on the pixel change is: ; In the formula, and is the weight coefficient; The degree of abnormal cell morphology; is the local contrast change rate.

[0028] Step S43: obtaining a cell characteristic value of each pixel point of the digital image according to the digital image pixel gradient value and the value of the influence factor of the region on the pixel change; The cell characteristic value is calculated as follows: ; In the formula, is the gradient value of the pixel point of the digital image; is the value of the influence factor of the region on the pixel change; and is the weight coefficient, and ; It should be noted that the gradient value is an important parameter for measuring the brightness change in an image. In a cell image, the area with a larger gradient value usually represents the edge or contour in the image; by calculating the gradient value of each pixel, the position of the cell edge and its change can be found; Cell feature value: The cell feature value of each pixel is calculated based on the gradient value of the pixel and the value of the influence factor of the region on the pixel change; this feature value represents a comprehensive feature of the cell area corresponding to the pixel, combining the morphological change of the area (through the degree of abnormality), the edge characteristics of the cell (through the gradient value) and the importance of the cell in the image (through the value of the influence factor of the region on the pixel change); Step S44: clustering is performed based on the cell feature values ​​to obtain a plurality of clusters; based on the clusters and the connection regions, the clusters and the connection regions are merged and supplemented in combination with a region growing algorithm to obtain a plurality of cell images; and the cell images are counted based on the cell images and the degree of abnormal cell morphology to obtain counting results.

[0029] It should be noted that the above-mentioned embodiment of the present application can effectively distinguish normal cells from abnormal cells and quantify the degree of abnormal morphology of cells by first dividing and analyzing the connected areas, and provide an important basis for subsequent cell counting and pathological analysis; further, by calculating the value of the influencing factor of the region on the pixel change, each pixel has a different influence on the cell morphology analysis, so that the area with large cell morphology changes (such as abnormal cells) can have a greater impact on the overall analysis results, thereby improving the accuracy of cell abnormality detection; further, in particular, by comprehensively considering multiple features (gradient values, local weights), a comprehensive feature value is provided for each pixel, and the accuracy of cell recognition is effectively improved, so that each pixel in the image is more accurately located in the cell morphology analysis; further, through cluster analysis, the cells in the image are effectively separated from the background and other non-cellular areas, providing a reliable basis for cell counting, and combined with the degree of abnormal cell morphology, normal cells and abnormal cells are further distinguished, and more accurate results are provided for cell counting.

[0030] Specifically, in step S44, based on the clusters and the connected regions, the clusters and the connected regions are merged and supplemented with a region growing algorithm to obtain a plurality of cell images, including the following steps: Step S441: acquiring a first cell image based on each of the clusters; It should be noted that the first cell image is constructed by the numerical image pixels corresponding to each cluster, that is, the digital image pixels corresponding to each cluster constitute a first cell image. However, in these first cell images, some cell feature values ​​have large deviations due to the clustering of the clusters, resulting in missing edges of some first cell images. Therefore, it is necessary to supplement their edges.

[0031] Step S442: performing region growing on edge pixel points corresponding to the first cell image to obtain a second cell image; Step S443: constructing a boundary point set based on the second edge pixel points of the second cell image; Step S444: Calculate and obtain the spatial fusion degree R of the boundary point set and the connection area; The calculation method of the spatial fusion degree R is: ; In the formula, A represents the set of boundary points; B represents the connected area; Step S445: Based on the spatial fusion degree R, the second cell image and the connection area are merged to obtain a cell image.

[0032] It should be noted that the above-mentioned embodiment of the present application first obtains a first cell image by clustering the pixel points in the bounding box of the cluster, and then, based on the edge missing and other conditions in the first cell image, performs regional growth processing on the edge pixel points to obtain a second cell image, so that the cell contour is clearer; further, by extracting and constructing a set of boundary points, the boundary of the cell is clearly defined, providing accurate input for the subsequent calculation of spatial fusion degree; then, by calculating the spatial fusion degree R between the boundary point set and the connection area (the clustering area in the first cell image), that is, measuring the similarity or consistency between the two, the degree of overlap between the boundary point set and the connection area is determined, so as to judge whether the boundary accurately matches the cell area; finally, according to the calculated spatial fusion degree R, the degree of fusion of the second cell image and the connection area is judged, and the images are merged based on the fusion degree to generate a more complete and accurate cell image. By referring to the spatial fusion degree, the merging process can minimize the error, ensure that the generated image is accurate in cell morphology and spatial distribution, and generate the final complete cell image.

[0033] Specifically, Figure 3 As shown, in step S442, region growing is performed on edge pixel points corresponding to the first cell image to obtain a second cell image, including the following operation steps: Step S4421: analyzing and calculating the first evaluation value and the second evaluation value after labeling the first cell image and the digital image; Step S4422: performing an iterative analysis region growing operation based on the first evaluation value and the second evaluation value combined with the edge pixel points to obtain a second cell image; It should be noted that if Figure 4 As shown, the bold lines and points in the figure are the edges of the unrecognized cell images; the above-mentioned embodiment of the present application first marks each pixel in the first cell image, distinguishes different areas, clarifies the cell area and background information, and can also identify edge pixels and non-edge pixels to avoid erroneous operations on the background or irrelevant areas in subsequent operations; Furthermore, by quantifying the relationship between edge pixels and other regions through the first judgment value and the second judgment value, it can be determined whether the edge pixels represent the contour of the cell accurately enough and whether further region growing processing is needed; then, through the region growing method, the region of the edge pixels of the first cell image is gradually expanded and merged to supplement the similarity of the edge pixels and their surroundings until a complete cell contour (i.e., the second cell image mentioned above) is formed, effectively filling the missing or incomplete regions of the cell edges caused by clustering deviations, especially for abnormal cells (such as medulloblastoma cells), the tumorous protrusions of the cells and other special morphologies can be better identified and represented, making the division of the cell region clearer and the edges smoother, thereby improving the quality of the cell image.

[0034] Specifically, in step S4421, the first evaluation value and the second evaluation value are obtained by analyzing and calculating the first cell image and the digital image after labeling, including the following operation steps: Step S44211: labeling each of the first cell images to obtain a cell image label; Step S44212: obtaining the cell image labels of the first cell pixels corresponding to each of the first cell images based on the cell image labels; Step S44213: based on all digital pixel points in the digital image, a plurality of unlabeled pixel points are obtained by screening out all the first cell pixel points; Step S44214: constructing an unlabeled pixel point set based on all the unlabeled pixel points; Step S44215: Calculate the first influencing evaluation factor based on the brightness value of each of the first cell pixels in the first cell image ; Calculate the second influencing factor based on the brightness value of the unlabeled pixel ; Step S44216: The first influencing evaluation factor is determined based on the brightness value of each of the first cell pixels in the first cell image. Calculate the first judgment value Based on the brightness value of the unlabeled pixel and the second influencing factor Calculate the second judgment value ; The first influencing factor The calculation method of (quantifying the brightness level of the cell area, thus reflecting the significance or quality of the cell) is: ; Where y is the brightness value of the first cell pixel; is the total number of pixels (including the first cell pixel and unlabeled pixels); The total number of pixels in the first cell; The second influencing factor The calculation method of is (reflecting the brightness characteristics of the background): ; Where x is the brightness value of the unlabeled pixel; e is the total number of unlabeled pixels; The calculation method of the first evaluation value is (reflecting the brightness deviation of the cell area and its consistency with the expected value): ; The second evaluation value is calculated as follows (reflecting the brightness deviation of the unlabeled area and its impact on the overall image quality): ; It should be noted that the above-mentioned embodiment of the present application first labels the cell image and assigns a unique label to the pixels in each cell image so that the area of ​​each cell can be tracked during subsequent processing, providing a clear basic framework for subsequent identification, analysis and segmentation, so that each cell area can be processed and analyzed separately, avoiding the mixing of different cell areas; further, each pixel point in the cell image is associated with the corresponding cell image label to ensure that the pixel point in each cell image can be associated with a specific cell label, so as to facilitate the accurate calculation of the illumination, morphological characteristics, etc. of the cell area, and avoid misjudgment or loss of information; further, by removing the cell pixels that have been labeled, the accurate distinction between the background and non-cell areas is guaranteed to avoid affecting the subsequent cell Boundary extension and morphological recognition; then, the screened unlabeled pixels are collected as a set, which provides a preliminary "candidate region" for the subsequent steps of region growing and cell boundary extension; further, by calculating the brightness values ​​of the first cell pixel and the unlabeled pixel, the first influencing judgment factor and the second influencing judgment factor are obtained, and the first influencing judgment factor and the second influencing judgment factor are used to reflect the brightness change of the cell area, and help determine which unlabeled pixels may belong to the cell area, providing a quantitative standard for the precise segmentation of the cell; finally, by combining the brightness value of the cell pixel and the influencing judgment factor, the first judgment value and the second judgment value are calculated, which provide a powerful judgment standard for the precise identification of morphologically abnormal cells (such as intracranial medulloblastoma cells).

[0035] Specifically, in step S4422, an iterative analysis region growing operation is performed based on the first evaluation value and the second evaluation value in combination with the edge pixel points to obtain a second cell image, including the following operation steps: Step S44221: performing a multi-factor analysis based on the coordinates of each digital image pixel point in the digital image and the brightness value of each digital image pixel point in combination with the edge pixel point to obtain an edge pixel point sequence set; traversing each edge pixel point in turn based on the order of each edge pixel point in the edge pixel point sequence set, and calculating a first local brightness contrast based on each unlabeled pixel point in the 8 neighborhood pixel points corresponding to the ti-th edge pixel point; Step S44222: Calculate the second local brightness contrast between the neighboring pixel corresponding to the cth unlabeled pixel in the current neighboring pixel and the tith edge pixel; and select a number of the unlabeled pixels less than or equal to the first local brightness contrast as candidate pixels based on the second local brightness contrast; Step S44223: determine whether the number a of the candidate pixels is greater than or equal to a preset minimum number threshold of the candidate pixels; if so (if not, it is determined to be an internal pixel of the cell image and the next processing operation is not performed), then the candidate pixel is used as a supplementary pixel to be determined; Step S44224: obtaining a new first cell image to be determined based on the supplementary pixel points to be determined and the first cell image; obtaining a new unlabeled pixel point set to be determined based on the new first cell image pixel points to be determined corresponding to the new first cell image to be determined and the digital image; Step S44225: Calculate the first judgment value change rate based on the brightness value of each pixel point of the new first cell image to be determined and the first judgment value in the new first cell image to be determined; calculate the second judgment value change rate based on the brightness value of each new unlabeled pixel point in the new unlabeled pixel point set to be determined and the second judgment value; Step S44226: determine whether the change rate of the first judgment value is less than the change rate of the second judgment value, if so (if not, it is determined to be a pixel point inside the cell image and the next processing operation is shelved), then determine the to-be-determined supplementary pixel point as a target supplementary pixel point, and merge the target supplementary pixel point into the first cell image to obtain a second cell image; and obtain a new unlabeled pixel point set based on the second cell image; update the second judgment value based on the brightness value of each second cell image pixel point in the second cell image or the brightness value of each new unlabeled pixel point in the new unlabeled pixel point set to obtain a new second judgment value; Step S44227: determine whether all edge pixel points have been traversed; if so, stop the iteration and output each of the second cell images; if not, return to the above step S44221 to continue the traversal operation until all edge pixel points have been traversed and multiple second cell images are output.

[0036] It should be noted that the above-mentioned embodiment of the present application firstly combines the coordinates, brightness values ​​and edge pixels of the digital image pixels, obtains the edge pixel sequence set through multi-factor analysis, determines the position of the edge pixels and sorts them, and provides a structured sequence for the subsequent cell boundary recognition and segmentation, which helps to gradually track the changes in cell boundaries, especially for cells with abnormal morphology (such as intracranial medulloblastoma cells with tumor-like protrusions); Furthermore, by calculating the local brightness contrast between the neighborhood pixels (8 neighborhoods) of the ti-th edge pixel and the unlabeled pixels one by one, the unlabeled pixels whose brightness contrast is less than or equal to the first local brightness contrast are screened out as candidate pixels. Through the analysis of the brightness contrast, the pixels located near the edge and having similar brightness characteristics are effectively identified, and the extended area of ​​the cell edge is identified, especially for the identification of the cell protrusion area with irregular morphology. Furthermore, the number of candidate pixels is judged. If the number of candidate pixels is greater than or equal to the preset minimum threshold, these pixels are considered to be target pixels to be supplemented, ensuring that the screening of candidate pixels is not over-filtered. , ensuring that important edge pixels are not missed; by gradually updating the cell image and the set of unlabeled pixels, the cell boundary can be expanded more finely and the segmentation result can be gradually improved; further, the accuracy of cell segmentation is reflected by calculating the judgment value change rate, ensuring that the boundary changes in the segmentation process are reasonable, and effectively avoiding excessive or missed segmentation, especially for cells with complex morphology (such as tumor-like protrusions). Then, by dynamically judging the judgment value change rate, the accuracy of the supplemented pixels is ensured, and the cell segmentation results are further optimized to ensure more accurate image segmentation, especially for those abnormal cells with complex morphology.

[0037] Specifically, in step S44221, a multi-factor analysis is performed based on the coordinates of each digital image pixel point in the digital image and the brightness value of each digital image pixel point in combination with the edge pixel point to obtain an edge pixel point sequence set, including the following operation steps: Step S442211: traverse each edge pixel point, and calculate the edge pixel point contrast coefficient F based on the brightness value of the edge pixel point and the neighborhood brightness values ​​of the neighboring pixels of the edge pixel point; The edge pixel contrast coefficient is calculated as follows: ; In the formula, is the brightness value of the edge pixel; is the average brightness value of the neighboring pixels of the edge pixel; G is the gradient value of the edge pixel; is the maximum gradient value in the digital image; It should be noted that the higher the contrast coefficient of the above-mentioned edge pixel point, the more it proves that the edge pixel point is the key position of the cell boundary. Prioritizing the key position of the cell boundary can process more obvious cell images more quickly, thereby screening out some unlabeled pixels that are easy to process (i.e., the edge missing part in the first cell image), thereby reducing the subsequent processing calculation pressure.

[0038] Step S442212: calculating the regional connectivity feature C based on the number of first cell pixels in the neighborhood pixels of the edge pixel, the number of window pixels within the preset window range of the edge pixel, and the number of first cell pixels in the window pixel; The calculation method of the regional connectivity feature C is: ; In the formula, is the number of first cell pixels in the neighborhood of edge pixels; is the number of the first cell pixel in the window pixel; The number of pixels in the window; It should be noted that in the above-mentioned embodiment of the present application, the larger the regional connectivity feature C is, the better the connectivity of the edge pixel point is, and the edge pixel points with better connectivity can be processed preferentially to ensure the connectivity of regional growth.

[0039] Step S442213: Obtain the coordinates of each of the first cell pixel points, and calculate the center point coordinates of each of the first cell images and the average radius R of the first cell image based on the coordinates of the first cell pixel points; calculate the center distance based on the coordinates of the edge pixel points and the center point coordinates of the first cell image corresponding to the edge pixel points; calculate the regional pixel density based on the number of first cell pixel points within the preset window range; calculate the distribution feature D based on the regional pixel density, the average radius R, the center distance and the preset target density; The distribution characteristic D is calculated as follows: ; Where d is the center distance; is the average radius; is the regional pixel density; is the target density; It should be noted that the larger the distribution feature D is, the more reasonable the distribution of cell pixels is. Edge pixels with more reasonable distribution can be processed preferentially to avoid excessive expansion or contraction of cells, thereby maintaining a reasonable distribution in space and making subsequent analysis and calculation more accurate.

[0040] Step S442214: Calculate the edge pixel processing priority L based on the edge pixel contrast coefficient F, the region connectivity feature C and the distribution feature D; The edge pixel processing priority L is calculated as follows: ; In the formula, α1, β1 and γ1 are weight coefficients; Step S442215: sorting the edge pixels from high to low based on the edge pixel processing priority L to obtain an edge pixel sequence set; It should be noted that the above-mentioned embodiment of the present application realizes efficient sorting of edge pixels by comprehensively considering the edge pixel contrast coefficient F, the region connectivity feature C and the distribution feature D, thereby accelerating the recognition and analysis speed of the second cell image.

[0041] In summary, the present invention proposes a method for detecting, identifying and counting cerebrospinal fluid cell images. A high-quality digital image of a cerebrospinal fluid cell smear is obtained through a microscope, and then the edges and other regions of the image for distinguishing cells are revealed by calculating the gradient value. Furthermore, the microscopic changes of the cell image are accurately captured by calculating the local contrast change rate, and normal cells and cells with abnormal morphology are effectively distinguished. Then, normal cells and abnormal cells are counted to obtain counting results, so as to better help medical staff identify the patient's condition and provide powerful patient medical record information for subsequent medical assistance. Among them, in particular, by comprehensively considering multiple features (gradient value, local weight), a comprehensive feature value is provided for each pixel point, which is combined with the degree of abnormal cell morphology to distinguish normal cells from abnormal cells and provide more accurate results for cell counting; In the specific operation process, clusters and connected regions are combined with the region growing algorithm to merge and supplement to generate the final complete cell image; During specific execution, the first judgment value and the second judgment value are comprehensively calculated by the brightness values ​​of the labeled pixels of the first cell image and the unlabeled pixels, and the second cell image is obtained by iteratively analyzing the regional growth operation in combination with the edge pixels. The edge pixel contrast coefficient F, the regional connectivity feature C and the distribution feature D are comprehensively considered to construct an edge pixel sequence set to accelerate the recognition and analysis speed of the second cell image. In combination with the local brightness contrast, the second cell image is gradually completed by dynamically judging the change rate of the judgment value, and the cell edge missing or incomplete area caused by the cluster deviation is effectively filled.

[0042] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. A person skilled in the art may modify the technical solutions described in the above embodiments, or replace part or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting, identifying and counting cerebrospinal fluid cell images, characterized in that: The steps are as follows: Acquire digital images of cerebrospinal fluid cell smears; Obtaining a digital image pixel gradient value corresponding to each digital image pixel according to a grayscale value difference between adjacent digital image pixels of each digital image pixel in the digital image of the cerebrospinal fluid cell smear; Analyze the local directional gradient in the preset neighborhood window of each digital image pixel to obtain the local contrast change rate of each digital image pixel; Based on the local contrast change rate combined with the degree of abnormal cell morphology and the digital image pixel gradient value, the cell image in the digital image is identified and counted to obtain a counting result; The counting result includes the number of normal cell types counted and the number of abnormal cell types counted.

2. The method for detecting, identifying and counting cerebrospinal fluid cell images according to claim 1, characterized in that: The method of identifying and counting the cell images in the digital image based on the local contrast change rate combined with the degree of abnormal cell morphology and the digital image pixel gradient value to obtain a counting result includes the following steps: Acquire a connection area of ​​digital image pixel points according to the local contrast change rate, and obtain the degree of abnormal cell morphology of the cerebrospinal fluid cell smear according to the distribution characteristics corresponding to the connection area; According to the degree of abnormal cell morphology and the local contrast change rate corresponding to each digital image pixel, the value of the influence factor of the region on the pixel change is obtained by analysis and calculation; Obtaining a cell characteristic value of each pixel point of the digital image according to the digital image pixel point gradient value and the value of the influence factor of the region on the pixel point change; Clustering is performed based on the cell feature values ​​to obtain a plurality of clusters; based on the clusters and the connection regions, the clusters and the connection regions are merged and supplemented in combination with a region growing algorithm to obtain a plurality of cell images; The cell images are counted based on the cell images and the degree of abnormal cell morphology to obtain a counting result.

3. The method for detecting, identifying and counting cerebrospinal fluid cell images according to claim 2, characterized in that: The calculation method of the value of the influence factor of the area on the pixel change is: ; In the formula, and is the weight coefficient; The degree of abnormal cell morphology; is the local contrast change rate; The cell characteristic value is calculated as follows: ; In the formula, is the gradient value of the pixel point of the digital image; is the value of the influence factor of the region on the pixel change; and is the weight coefficient, and .

4. The method for detecting, identifying and counting cerebrospinal fluid cell images according to claim 3, characterized in that: The step of merging and supplementing the clusters and the connected regions based on the clusters and the connected regions in combination with a region growing algorithm to obtain a plurality of cell images includes the following steps: Acquire a first cell image based on each of the clusters; Performing region growing on edge pixel points corresponding to the first cell image to obtain a second cell image; Constructing a boundary point set based on second edge pixel points of the second cell image; Calculate and obtain the spatial fusion degree R of the boundary point set and the connection area; The second cell image and the connection area are merged based on the spatial fusion degree R to obtain a cell image.

5. The method for detecting, identifying and counting cerebrospinal fluid cell images according to claim 4, characterized in that: The step of performing region growing on edge pixels corresponding to the first cell image to obtain a second cell image comprises the following steps: Based on the first cell image and the digital image, the first evaluation value and the second evaluation value are obtained by analyzing and calculating after labeling; An iterative analysis region growing operation is performed based on the first judgment value and the second judgment value in combination with the edge pixel points to obtain a second cell image.

6. The method for detecting, identifying and counting cerebrospinal fluid cell images according to claim 5, characterized in that: The step of analyzing and calculating the first evaluation value and the second evaluation value after labeling the first cell image and the digital image comprises the following steps: Labeling each of the first cell images to obtain a cell image label; Obtaining cell image labels of first cell pixels corresponding to each of the first cell images based on the cell image labels; Based on all digital pixels in the digital image, a plurality of unlabeled pixels are obtained by removing all the first cell pixels; Constructing an unlabeled pixel point set based on all the unlabeled pixel points; The first influencing evaluation factor is obtained by calculating based on the brightness value of each of the first cell pixels in the first cell image. ; Calculate the second influencing factor based on the brightness value of the unlabeled pixel ; The first influencing evaluation factor is based on the brightness value of each of the first cell pixels in the first cell image. Calculate the first judgment value ; Based on the brightness value of the unlabeled pixel and the second influencing evaluation factor Calculate the second judgment value .

7. The method for detecting, identifying and counting cerebrospinal fluid cell images according to claim 6, characterized in that: The first influencing factor The calculation method is: ; Where y is the brightness value of the first cell pixel; is the total number of pixels; The total number of pixels in the first cell; The second influencing factor The calculation method is: ; Where x is the brightness value of the unlabeled pixel; e is the total number of unlabeled pixels; The calculation method of the first evaluation value is: ; The second evaluation value is calculated as follows: .

8. The method for detecting, identifying and counting cerebrospinal fluid cell images according to claim 7, characterized in that: The iterative analysis of the region growing operation based on the first evaluation value and the second evaluation value combined with the edge pixel points to obtain the second cell image includes the following operation steps: Based on the coordinates of each digital image pixel point in the digital image and the brightness value of each digital image pixel point, a multi-factor analysis is performed in combination with the edge pixel points to obtain an edge pixel point sequence set; based on the order of each edge pixel point in the edge pixel point sequence set, each edge pixel point is sequentially traversed, and a first local brightness contrast is calculated based on each unlabeled pixel point in the 8 neighboring pixel points corresponding to the ti-th edge pixel point; Calculate the second local brightness contrast between the neighboring pixel point corresponding to the cth unlabeled pixel point in the current neighboring pixel points and the tith edge pixel point; and select a number of the unlabeled pixel points less than or equal to the first local brightness contrast as candidate pixel points based on the second local brightness contrast; Determine whether the number a of the candidate pixel points is greater than or equal to a preset minimum number threshold of the candidate pixel points; if so, use the candidate pixel points as supplementary pixel points to be determined; Obtaining a new first cell image to be determined based on the supplementary pixel points to be determined and the first cell image; Obtaining a new unlabeled pixel point set to be determined based on the new first cell image pixel points to be determined corresponding to the new first cell image to be determined and the digital image; The first judgment value change rate is calculated based on the brightness value of each pixel point of the new first cell image to be determined in the new first cell image to be determined and the first judgment value; The second judgment value change rate is calculated based on the brightness value of each new unlabeled pixel in the set of new unlabeled pixel points to be determined and the second judgment value; Determine whether the first evaluation value change rate is less than the second evaluation value change rate, and if so, determine the to-be-determined supplementary pixel point as a target supplementary pixel point, and merge the target supplementary pixel point into the first cell image to obtain a second cell image; and obtain a new unlabeled pixel point set based on the second cell image; The second evaluation value is updated based on the brightness value of each second cell image pixel in the second cell image or the brightness value of each new unlabeled pixel in the new unlabeled pixel set to obtain a new second evaluation value; Determine whether all edge pixel points have been traversed; if so, stop the iteration and output each of the second cell images; if not, return to the above step to continue the traversal operation until all edge pixel points have been traversed and multiple second cell images are output.

9. The method for detecting, identifying and counting cerebrospinal fluid cell images according to claim 8, characterized in that: The step of performing a multi-factor analysis based on the coordinates of each digital image pixel point in the digital image and the brightness value of each digital image pixel point in combination with the edge pixel point to obtain an edge pixel point sequence set includes the following steps: Traversing each of the edge pixels, and calculating the edge pixel contrast coefficient F based on the brightness value of the edge pixel and the neighborhood brightness values ​​of the neighborhood pixels of the edge pixel; Calculating a regional connectivity feature C based on the number of first cell pixels in the neighborhood pixels of the edge pixel, the number of window pixels within a preset window range of the edge pixel, and the number of first cell pixels in the window pixel; Obtaining the coordinates of each of the first cell pixel points, and calculating the coordinates of the center point of each of the first cell images and the average radius R of the first cell images based on the coordinates of the first cell pixel points; and calculating the center distance based on the coordinates of the edge pixel points and the coordinates of the center point of the first cell image corresponding to the edge pixel points; Calculating the regional pixel density based on the number of first cell pixels within the preset window range; The distribution feature D is calculated based on the pixel density of the region, the average radius R, the center distance and the preset target density; The edge pixel processing priority L is calculated based on the edge pixel contrast coefficient F, the region connectivity feature C and the distribution feature D; The edge pixel points are sorted from high to low based on the edge pixel point processing priority L to obtain an edge pixel point sequence set.

Images