Dyeing cell positioning method and system for pathological diagnosis

Abstract

The invention discloses a staining cell positioning method and system for pathological diagnosis, and relates to the technical field of staining cell positioning. The staining cell positioning method for pathological diagnosis comprises the following steps: acquiring an original pathological staining image based on an image acquisition device, and preprocessing the original pathological staining image to obtain a processed pathological staining image; feature processing is conducted on the processed pathological staining image, a staining area binary image and staining cell feature data of pixel points are obtained, and the staining cell feature data comprise shape features, texture features and color features; carrying out image segmentation on the dyed area binary image based on the dyed cell characteristic data of the pixel points to obtain a segmented pathological dyed image; and cell localization is carried out based on the segmented pathological staining image, so that the problem that the existing staining cell localization method is inconvenient to effectively separate cells by combining region growth and watershed algorithms is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of stained cell localization, and specifically to a stained cell localization method and system for pathological diagnosis. Background Art

[0002] Cell staining is a technical means widely used in clinical work and scientific research. It reflects the cell morphology and the expression of molecular markers of the sample to be tested through staining information, and can provide important decision-making information for clinicians and scientific researchers. Among them, cell localization and cell segmentation are of great significance in cell classification, counting, and analysis of staining results.

[0003] The Chinese invention patent with the publication number CN108074243B discloses a cell localization method and a cell segmentation method. Machine learning is performed on a first stained image containing cell localization information to obtain a prediction model. The prediction model is used for a second stained image that does not contain cell localization information to predict the cell localization information of the second stained image, and cell segmentation is performed on the second stained image based on the cell localization information.

[0004] However, the existing stained cell localization methods are inconvenient to combine the region growing and watershed algorithms to effectively separate cells. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a stained cell localization method and system for pathological diagnosis, which solves the problem that the existing stained cell localization methods are inconvenient to combine the region growing and watershed algorithms to effectively separate cells.

[0006] To achieve the above object, the present invention is realized through the following technical solutions: A stained cell localization method for pathological diagnosis includes the following steps: acquiring an original pathological stained image based on an image acquisition device and performing preprocessing to obtain a processed pathological stained image; performing feature processing on the processed pathological stained image to obtain a binary image of the stained region and the stained cell feature data of the pixel points, where the stained cell feature data includes shape features, texture features, and color features; performing image segmentation on the binary image of the stained region based on the stained cell feature data of the pixel points to obtain a segmented pathological stained image; and performing cell localization based on the segmented pathological stained image.

[0007] Further, obtaining the binary image of the stained area and the characteristic data of stained cells includes the following steps: converting the processed pathological stained image from the RGB color space to the HSV color space to obtain the HSV stained image; performing binary segmentation on the HSV stained image based on the threshold segmentation algorithm to obtain the binary image of the stained area; extracting the shape features, texture features, and color features of the stained area segmented from the binary image of the stained area to obtain the characteristic data of stained cells of pixel points.

[0008] Further, the shape features include area, circularity, and aspect ratio; the texture features include gray-level co-occurrence matrix energy, gray-level co-occurrence matrix entropy, gray-level co-occurrence matrix contrast, and gray-level co-occurrence matrix correlation; the color features include hue, saturation, and value.

[0009] Further, performing image segmentation on the binary image of the stained area based on the characteristic data of stained cells of pixel points to obtain the segmented pathological stained image includes the following steps: constructing a similarity measurement condition, and performing preliminary segmentation on the binary image of the stained area by combining region growing to obtain a preliminarily segmented cell image; optimizing the preliminarily segmented cell image based on the watershed algorithm to obtain the segmented pathological stained image.

[0010] Further, performing preliminary segmentation on the binary image of the stained area by combining region growing to obtain a preliminarily segmented cell image includes the following steps: in the binary image of the stained area, traverse all initial pixel points with a pixel value of 1, and randomly select a set number of initial pixel points from them as the seed point set S; create an empty set Q of pixels to be processed, and add all the seed points s in the seed point set S to the set Q of pixels to be processed; when the set Q of pixels to be processed is not empty, take out one of the pixels to be processed, and traverse its 8-neighborhood pixel points. For each neighborhood pixel point p, if the pixel value of the neighborhood pixel point p in the binary image of the stained area is 0 and simultaneously satisfies the similarity measurement condition, then set the pixel value of the neighborhood pixel point p to 1 and add it to the set Q of pixels to be processed; repeat the above process until the set Q of pixels to be processed is empty to obtain the preliminarily segmented cell image.

[0011] Furthermore, the process of constructing the similarity measurement condition is as follows: Obtain the stained cell feature data of a pixel point and the actual stained cell feature data of its neighboring pixel points. The actual stained cell feature data includes the actual shape feature, the actual texture feature, and the actual color feature, where: The actual shape feature includes the actual area, the actual circularity, and the actual aspect ratio; The actual texture feature includes the actual energy of the gray-level co-occurrence matrix, the actual entropy of the gray-level co-occurrence matrix, the actual contrast of the gray-level co-occurrence matrix, and the actual correlation of the gray-level co-occurrence matrix; The actual color feature includes the actual hue, the actual saturation, and the actual brightness; Construct a shape similarity function based on the shape feature and the actual shape feature to determine the shape similarity value; Construct a texture similarity function based on the texture feature and the actual texture feature to determine the texture similarity value; Construct a color similarity function based on the color feature and the actual color feature to determine the color similarity value; The similarity measurement condition is: the shape similarity value is not greater than the shape similarity threshold, the texture similarity value is not greater than the texture similarity threshold, and the color similarity value is not greater than the color similarity threshold.

[0012] Furthermore, the shape similarity function is: ; Wherein, is the area of the stained region where the seed point s is located, is the actual area of the stained region where the neighboring pixel point p is located, is 's weight factor, is the circularity of the stained region where the seed point s is located, is the actual circularity of the stained region where the neighboring pixel point p is located, is 's weight factor, is the aspect ratio of the stained region where the seed point s is located, is the actual aspect ratio of the stained region where the neighboring pixel point p is located, is 's weight factor; The texture similarity function is: ; Wherein, is the energy of the gray-level co-occurrence matrix of the stained region where the seed point s is located, is the actual energy of the gray-level co-occurrence matrix of the stained region where the neighboring pixel point p is located, is the entropy of the gray-level co-occurrence matrix of the stained region where the seed point s is located, is the actual entropy of the gray-level co-occurrence matrix of the stained region where the neighboring pixel point p is located, is the contrast of the gray-level co-occurrence matrix of the stained region where the seed point s is located, is the actual contrast of the gray-level co-occurrence matrix of the staining region where the neighborhood pixel point p is located, is the correlation of the gray-level co-occurrence matrix of the staining region where the seed point s is located, the actual correlation of the gray-level co-occurrence matrix of the staining region where the neighborhood pixel point p is located; Color similarity function is: ; where, are the coordinates of the seed point s in the HSV color space, representing hue, saturation, and lightness respectively, are the coordinates of the neighborhood pixel point p in the HSV color space, representing the actual hue, actual saturation, and actual lightness respectively.

[0013] Furthermore, the preliminary segmented cell image is optimized based on the watershed algorithm to obtain the segmented pathological staining image, including the following steps: Calculate the gradients of the preliminary segmented cell image in the horizontal and vertical directions based on the Sobel operator. For each pixel point, the horizontal gradient and the vertical gradient are calculated as follows: ; ; where, is the gray value of the original pathological staining image at the pixel point , represents the convolution operation, and calculates the gradient magnitude of each pixel point, ; Traverse each cell region in the preliminary segmented cell image to determine the region area and the region circularity ; If the region circularity is greater than the set circularity threshold and less than 1, and at the same time the region area satisfies , then multiply the gradient magnitude at the boundary of the cell region by a first correction coefficient, and the first correction coefficient is less than 1, where, is the set minimum region area value, is the set maximum region area value; If the region circularity is greater than 1, and the region area satisfies , then multiply the gradient magnitude at the boundary of the cell region by a second correction coefficient, and the second correction coefficient is less than 1; Obtain the corrected gradient map; Perform a watershed transformation on the corrected gradient map to obtain the segmented pathological staining image.

[0014] Further, perform cell localization based on the segmented pathological staining image, including the following steps: Determine the segmented cell regions in the segmented pathological staining image, calculate the geometric center coordinates of each segmented cell region, and denote them as cell center coordinates ; ; where is the coordinate of the a-th pixel point in the segmented cell region, and n is the total number of pixel points in the segmented cell region; Mark the cell center coordinates based on the set marking type.

[0015] A staining cell localization system for pathological diagnosis, which is used for the above-mentioned staining cell localization method for pathological diagnosis, includes: a preprocessing module, which acquires the original pathological staining image based on an image acquisition device and performs preprocessing to obtain a processed pathological staining image; a feature extraction module, which performs feature processing on the processed pathological staining image to obtain a binary image of the staining region and staining cell feature data of pixel points, and the staining cell feature data includes shape features, texture features, and color features; a segmentation module, which performs image segmentation on the binary image of the staining region based on the staining cell feature data of pixel points to obtain a segmented pathological staining image; a localization module, which performs cell localization based on the segmented pathological staining image.

[0016] The present invention has the following beneficial effects: The staining cell localization method for pathological diagnosis can improve the image quality by acquiring the original pathological staining image and performing preprocessing; extract staining cell feature data from multiple dimensions to accurately distinguish cells; perform image segmentation based on these features, combining region growing and watershed algorithms, effectively separating cells and avoiding misjudgment due to adhesion; finally, accurately locate the cells, calculate the center coordinates and mark them, which is convenient for identification and analysis.

[0017] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flowchart of the staining cell localization method for pathological diagnosis of the present invention.

[0019] Figure 2 It is a flowchart block diagram of the staining cell localization system for pathological diagnosis of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] Please refer to Figure 1, the embodiments of the present invention provide a technical solution: a staining cell localization method for pathological diagnosis, including the following steps: obtaining an original pathological staining image based on an image acquisition device and performing preprocessing to obtain a processed pathological staining image; Ensure that the image is clear and has a high enough resolution to clearly show the morphology and staining characteristics of the cells. The acquired image format is common ones such as TIFF, JPEG, etc. Use the histogram equalization method to process the original image to enhance the contrast of the image, making the boundaries between cells and the background and between different cells clearer. The specific operation is to calculate the grayscale histogram of the image, redistribute the pixel grayscale values according to the histogram distribution, and expand the grayscale dynamic range of the image. Use the median filter algorithm to denoise the enhanced image. Median filtering replaces the grayscale value of each pixel point in the image with the median of the pixel grayscale values in its neighborhood, effectively removing impulse noises such as salt-and-pepper noise, and at the same time better retaining the edge information of the image, avoiding noise interference in the subsequent extraction and analysis of cell characteristics.

[0021] Perform feature processing on the processed pathological staining image to obtain a binary image of the stained area and the stained cell feature data of the pixel points. The stained cell feature data includes shape features, texture features, and color features.

[0022] Convert the processed pathological staining image from the RGB color space to the HSV color space to obtain an HSV staining image; the RGB color space focuses on color mixing, while the HSV color space describes colors from three dimensions that more intuitively reflect human perception of colors: hue, saturation, and value. In pathological diagnosis, the colors of different cells after staining can more prominently show their characteristic differences in the HSV space. For example, the hue changes of some diseased cells after staining are more easily captured in the HSV space, which is more conducive to the subsequent analysis of the staining cell characteristics compared to the RGB space, improving the pertinence of the analysis of the color characteristics of staining cells.

[0023] Perform binary segmentation on the HSV staining image based on the threshold segmentation algorithm to obtain a binary image of the stained area; the threshold segmentation algorithm can divide the image pixels into two categories (usually foreground and background) according to the set threshold, separating the stained cell area from the background to form a binary image. In pathological images, complex backgrounds and cell morphologies are likely to interfere with cell localization. After binarization, the stained cell area is presented with a unified pixel value (such as 1 representing the stained area and 0 representing the background), greatly simplifying the recognition of the cell area in the subsequent processing process, making the subsequent feature extraction and localization operations for stained cells easier to perform, and improving the processing efficiency and accuracy.

[0024] Shape feature extraction, texture feature extraction, and color feature extraction are performed on the segmented stained regions in the binary image of the stained region to obtain the stained cell feature data of pixel points. Shape features (area, circularity, and aspect ratio) can be used to distinguish different types of cells. For example, there are differences in shape between cancer cells and normal cells. Cancer cells may have a more irregular shape. By analyzing these shape features, the category and state of the cells can be preliminarily judged. Texture features (energy, entropy, contrast, and correlation of the gray-level co-occurrence matrix) reflect the characteristics such as the coarseness and complexity of the internal texture of the cells. The cell texture will be different under different pathological states, which helps to further distinguish normal and abnormal cells. Color features (hue, saturation, and lightness) can reflect the differences in staining. Cells stained with different stains have obvious differences in color features, providing an important basis for cell recognition in the color dimension. Combining these multi-dimensional feature data provides a comprehensive and reliable basis for subsequent accurate image segmentation and cell localization, improving the accuracy and reliability of cell localization.

[0025] Shape features include area, circularity, and aspect ratio; texture features include energy of the gray-level co-occurrence matrix, entropy of the gray-level co-occurrence matrix, contrast of the gray-level co-occurrence matrix, and correlation of the gray-level co-occurrence matrix; color features include hue, saturation, and lightness.

[0026] Based on the stained cell feature data of pixel points, image segmentation is performed on the binary image of the stained region to obtain the segmented pathological stained image, including the following steps: constructing a similarity measurement condition, and combining region growing to perform preliminary segmentation on the binary image of the stained region to obtain a preliminary segmented cell image; constructing the similarity measurement condition is based on the stained cell feature data such as the shape, texture, and color of pixel points to measure the similarity between neighboring pixel points and the seed points. The region growing algorithm starts with the seed points and gradually merges similar neighboring pixel points according to the similarity measurement condition. This process can effectively utilize the multi-feature information of cells, divide pixel points with similar features into the same cell region, and initially separate each cell. In pathological images, the features of different cells are different. This multi-feature-based region growing segmentation method can better adapt to the complexity of cell morphology and features, avoid mismerging different cells, improve the accuracy of segmentation, and lay a good foundation for subsequent accurate cell localization.

[0027] Optimize the preliminarily segmented cell image based on the watershed algorithm to obtain the segmented pathological stained image. The watershed algorithm simulates the flow of water on the terrain to segment the image. There may be some inaccurate boundaries or adhesion regions in the preliminarily segmented cell image. By calculating the gradient based on the Sobel operator, the boundary information of the cell region can be highlighted. For cell regions of different shapes and sizes, the boundary gradient values are corrected according to their circularity and area, making the cell boundaries more prominent and accurate in the gradient map. Then, the watershed transformation is performed, which can more precisely separate adjacent cells on the basis of the preliminary segmentation, avoid cell adhesion, and further refine the boundaries of the cell regions, thus obtaining a more accurate segmented pathological stained image, greatly improving the accuracy of cell localization and ensuring the accurate determination of the position of each cell during subsequent localization.

[0028] In the binary image of the stained area, traverse all initial pixel points with a pixel value of 1, and randomly select a set number of initial pixel points from them as the seed point set S; Random selection can to a certain extent avoid the problem of one-sided segmentation results caused by overly concentrated selection positions, ensuring the randomness and representativeness of the distribution of seed points in the stained area. These seed points are the starting points of region growing, and their reasonable distribution helps to evenly grow cell regions subsequently, avoiding missing some cell parts or wrongly merging different cells, thereby improving the accuracy of segmentation and laying a foundation for accurately locating stained cells.

[0029] Create an empty set Q of pixels to be processed, and add all the seed points s in the seed point set S to the set Q of pixels to be processed; The set Q of pixels to be processed is like a "task queue", which orderly stores the pixel points that need to be processed, ensuring that each seed point and its subsequent eligible neighborhood pixel points can be processed in sequence, ensuring the orderly progress of the region growing process, avoiding chaos and omission in pixel point processing, and enabling the growth and segmentation of cell regions to gradually unfold according to the predetermined rules.

[0030] When the set Q of pixels to be processed is not empty, take out one of the pixels to be processed, and traverse its 8-neighborhood pixel points. For each neighborhood pixel point p, if the pixel value of the neighborhood pixel point p is 0 in the binary image of the stained area and at the same time meets the similarity measurement condition, then set the pixel value of the neighborhood pixel point p to 1 and add it to the set Q of pixels to be processed; Repeat the above process until the set Q of pixels to be processed is empty to obtain the preliminarily segmented cell image. The similarity measurement condition constructed based on the characteristic data of stained cells of pixel points can accurately judge whether the neighborhood pixel point belongs to the same cell region as the currently processed pixel point. By continuously incorporating eligible neighborhood pixel points into the current region, a complete cell region is gradually grown, effectively distinguishing different cells, improving the accuracy of cell segmentation, and providing a reliable segmented image for accurately locating stained cells subsequently.

[0031] The process of constructing the similarity measurement condition is as follows: Obtain the stained cell feature data of a pixel point and the actual stained cell feature data of its neighboring pixel points. The actual stained cell feature data includes actual shape features, actual texture features, and actual color features, where: The actual shape features include actual area, actual circularity, and actual aspect ratio; The actual texture features include actual energy of the gray-level co-occurrence matrix, actual entropy of the gray-level co-occurrence matrix, actual contrast of the gray-level co-occurrence matrix, and actual correlation of the gray-level co-occurrence matrix; The actual color features include actual hue, actual saturation, and actual lightness; Construct a shape similarity function based on the shape features and the actual shape features to determine the shape similarity value; Construct a texture similarity function based on the texture features and the actual texture features to determine the texture similarity value; Construct a color similarity function based on the color features and the actual color features to determine the color similarity value; The similarity measurement condition is: The shape similarity value is not greater than the shape similarity threshold, the texture similarity value is not greater than the texture similarity threshold, and the color similarity value is not greater than the color similarity threshold.

[0032] Obtain the stained cell feature data of a pixel point and the actual stained cell feature data of its neighboring pixel points, covering multiple dimensions such as shape, texture, and color. These rich data comprehensively reflect the feature information of the region where the pixel point and its neighboring pixel points are located. In pathological image analysis, there are differences in these features of different cells. The multi-dimensional data provides a comprehensive basis for accurately judging the similarity between pixel points and avoids misjudgment caused by relying on only a single feature.

[0033] Shape similarity function is: ; where is the area of the stained region where the seed point s is located, is the actual area of the stained region where the neighboring pixel point p is located, is 's weight factor, is the circularity of the stained region where the seed point s is located, is the actual circularity of the stained region where the neighboring pixel point p is located, is 's weight factor, is the aspect ratio of the stained region where the seed point s is located, is the actual aspect ratio of the stained region where the neighboring pixel point p is located, is 's weight factor; Texture similarity function is: ; where is the energy of the gray-level co-occurrence matrix of the stained region where the seed point s is located, is the actual energy of the gray-level co-occurrence matrix of the stained region where the neighborhood pixel point p is located, is the entropy of the gray-level co-occurrence matrix of the stained region where the seed point s is located, is the actual entropy of the gray-level co-occurrence matrix of the stained region where the neighborhood pixel point p is located, is the contrast of the gray-level co-occurrence matrix of the stained region where the seed point s is located, is the actual contrast of the gray-level co-occurrence matrix of the stained region where the neighborhood pixel point p is located, is the correlation of the gray-level co-occurrence matrix of the stained region where the seed point s is located, the actual correlation of the gray-level co-occurrence matrix of the stained region where the neighborhood pixel point p is located; Color similarity function is: ; where, are the coordinates of the seed point s in the HSV color space, representing hue, saturation, and value respectively, are the coordinates of the neighborhood pixel point p in the HSV color space, representing the actual hue, actual saturation, and actual value respectively.

[0034] Shape similarity functions, texture similarity functions, and color similarity functions are constructed respectively based on shape features and actual shape features, texture features and actual texture features, and color features and actual color features, and then shape similarity values, texture similarity values, and color similarity values are determined. By quantifying the similarity degree of different features between pixel points in the form of mathematical functions, the judgment of pixel point similarity becomes more accurate and scientific. For example, the shape similarity function comprehensively considers shape factors such as area, circularity, and aspect ratio, and can more accurately measure the similarity degree of the shapes of two regions, which helps to accurately judge whether neighborhood pixel points should be grouped into the same cell region during the region growing process.

[0035] Taking the shape similarity value not greater than the color similarity threshold, the texture similarity value not greater than the texture similarity threshold, and the color similarity value not greater than the shape similarity threshold as the similarity measurement conditions. This way of integrating multiple features and setting thresholds ensures that during region growing, only pixel points that meet a certain similarity degree in multiple feature dimensions will be merged into the same region. In complex pathological images, it can effectively avoid mis-merging pixel points belonging to different cells, improve the accuracy of cell region segmentation, and thus provide a more reliable basis for subsequent accurate cell localization.

[0036] Based on the stained cell feature data of pixel points, image segmentation is performed on the binary image of the stained region to obtain the segmented pathological stained image; Optimizing the preliminarily segmented cell image based on the watershed algorithm to obtain the segmented pathological stained image, including the following steps: Calculating the gradients of the preliminarily segmented cell image in the horizontal and vertical directions based on the Sobel operator. For each pixel point, the horizontal direction gradient and the vertical direction gradient , and the calculation formulas are as follows: ; ; where is the gray value of the original pathological stained image at the pixel point , represents the convolution operation, calculating the gradient magnitude of each pixel point, ; Calculate the gradients of the preliminarily segmented cell image in the horizontal and vertical directions using the Sobel operator, and further calculate the gradient magnitude. In the pathological image, the pixel gray value changes significantly at the cell boundary. By calculating the gradient, these boundary information can be effectively highlighted. For example, at the boundary between normal cells and diseased cells, the pixel gray value change will be presented as a relatively large gradient magnitude after gradient calculation, making the cell boundary clearer and providing crucial boundary clues for subsequent accurate cell region division.

[0037] Traverse each cell region in the preliminarily segmented cell image to determine the region area and the region circularity ( , is the aspect ratio, is the pi); If the region circularity is greater than the set circularity threshold and less than 1, and at the same time the region area satisfies , then the gradient magnitude at the boundary of this cell region is multiplied by a first correction factor, and the first correction factor is less than 1, where is the set minimum region area value, is the set maximum region area value; If the region circularity is greater than 1, and the region area satisfies , then the gradient magnitude at the boundary of this cell region is multiplied by a second correction factor, and the second correction factor is less than 1; Obtain the corrected gradient map; Traverse each cell region of the preliminarily segmented cell image, and correct the gradient values of the boundaries of the cell regions according to the regional circularity and area. When the regional circularity is greater than the set circularity threshold and less than 1, or greater than 1 and the regional area meets the corresponding conditions, multiply by a correction factor less than 1. This is because for cell regions of different shapes and sizes, the boundary gradient situations are different. For example, for irregularly shaped cell regions, there may be interference information in the boundary gradient. By correction, these interferences can be weakened, highlighting the true cell boundary gradient features, avoiding incorrect segmentation during the watershed transformation, and improving the accuracy of cell region segmentation.

[0038] Perform a watershed transformation on the corrected gradient map to obtain the segmented pathological stained image.

[0039] First, regard each pixel point in the corrected gradient map as a point on the terrain surface, and the gray value of the pixel corresponds to the height of the point. By simulating the flow of water on the terrain, start filling from different "valleys" (i.e., low-gradient regions inside the cells) and stop at the "ridges" (i.e., high-gradient regions of the cell boundaries). The specific implementation can use a marker-based watershed algorithm. First, perform threshold segmentation on the corrected gradient map to obtain foreground markers and background markers, and then perform a watershed transformation based on the markers to accurately segment each cell and obtain a more accurate segmented pathological stained image.

[0040] Perform cell localization based on the segmented pathological stained image.

[0041] Determine the segmented cell regions in the segmented pathological stained image, calculate the geometric center coordinates of each segmented cell region, and denote them as cell center coordinates ; ; where, is the coordinate of the ath pixel point in the segmented cell region, n is the total number of pixel points in the segmented cell region; determine the segmented cell regions in the segmented pathological stained image, and calculate the geometric center coordinates of each region as the cell center coordinates. In pathological diagnosis, it is crucial to clarify the central position of the cells. For example, when analyzing the cell distribution pattern and judging whether the cell aggregation is abnormal, the cell center coordinates are key data. By accurately calculating the cell center coordinates, it is possible to accurately know the position of each cell in the image, providing basic positioning information for subsequent pathological analysis and helping doctors quickly and accurately observe and evaluate the cell distribution state.

[0042] Mark the cell center coordinates based on the set marker type.

[0043] A stained cell localization system for pathological diagnosis, for the above-mentioned stained cell localization method for pathological diagnosis, such as Figure 2As shown in the figure, it includes: a preprocessing module, which acquires an original pathological stained image based on an image acquisition device and performs preprocessing to obtain a processed pathological stained image; a feature extraction module, which performs feature processing on the processed pathological stained image to obtain a binary image of the stained area and stained cell feature data of pixel points, and the stained cell feature data includes shape features, texture features, and color features; a segmentation module, which performs image segmentation on the binary image of the stained area based on the stained cell feature data of pixel points to obtain a segmented pathological stained image; and a localization module, which performs cell localization based on the segmented pathological stained image.

[0044] An electronic device includes: a processor; and a memory, in which computer program instructions are stored, and when the computer program instructions are run by the processor, the processor is caused to execute the method for staining cell localization for pathological diagnosis as described above.

[0045] A computer-readable storage medium is used to store a program, and when the program is executed by a processor, the method for staining cell localization for pathological diagnosis as described above is implemented.

[0046] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0047] The present invention is described with reference to the flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0048] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in Figure 1 one process or multiple processes and / or blocksFigure 1 The functions specified in one or more boxes.

[0049] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are performed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 process or more processes and / or boxes Figure 1 or more boxes.

[0050] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0051] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A staining cell localization method for pathological diagnosis, characterized in that: The following steps are involved: Acquire the original pathological staining image based on the image acquisition device, and perform preprocessing to obtain the processed pathological staining image; Perform feature processing on the processed pathological staining image to obtain the binary image of the staining area and the staining cell feature data of the pixel points, the staining cell feature data including shape features, texture features and color features; Perform image segmentation on the binary image of the stained area based on the characteristic data of the stained cells at the pixel points to obtain a segmented pathological staining image; Cell localization was performed based on segmented pathological staining images.

2. The method for localizing stained cells for pathological diagnosis according to claim 1, characterized in that: Obtaining the binary image of the stained area and the characteristic data of the stained cells includes the following steps: Convert the processed pathological staining image from RGB color space to HSV color space to obtain an HSV staining image; The HSV stained image is segmented into two values ​​based on the threshold segmentation algorithm to obtain a binary image of the stained area; Shape feature extraction, texture feature extraction and color feature extraction are performed on the stained area segmented from the stained area binary image to obtain the stained cell feature data of the pixel points.

3. The method for localizing stained cells for pathological diagnosis according to claim 2, characterized in that: Shape features include area, circularity, and aspect ratio; Texture features include gray-level co-occurrence matrix energy, gray-level co-occurrence matrix entropy, gray-level co-occurrence matrix contrast and gray-level co-occurrence matrix correlation; Color characteristics include hue, saturation, and value.

4. The method for localizing stained cells for pathological diagnosis according to claim 1, characterized in that: The image segmentation of the stained region binary image is performed based on the stained cell feature data of the pixel points to obtain a segmented pathological stained image, including the following steps: Construct similarity measurement conditions, combine region growing to perform preliminary segmentation on the stained region binary image, and obtain preliminary segmented cell images; The preliminary segmented cell image was optimized based on the watershed algorithm to obtain the segmented pathological staining image.

5. The method for localizing stained cells for pathological diagnosis according to claim 4, characterized in that: The stained region binary image is preliminarily segmented based on region growing to obtain a preliminarily segmented cell image, including the following steps: In the binary image of the stained area, all initial pixel points with a pixel value of 1 are traversed, and a set number of initial pixel points are randomly selected as the seed point set S; Create an empty pixel set Q to be processed, and add all seed points s in the seed point set S to the pixel set Q to be processed; When the set of pixels to be processed Q is not empty, take out one of the pixels to be processed and traverse its 8 neighboring pixels. For each neighboring pixel p, if the pixel value of the neighboring pixel p in the stained area binary image is 0 and satisfies the similarity measurement condition, then set the pixel value of the neighboring pixel p to 1 and add it to the set of pixels to be processed Q; The above process is repeated until the pixel set Q to be processed is empty, and a preliminary segmented cell image is obtained.

6. The method for localizing stained cells for pathological diagnosis according to claim 5, characterized in that: The similarity measurement condition construction process is as follows: The stained cell feature data of the pixel point and the actual stained cell feature data of the neighboring pixel points are obtained. The actual stained cell feature data includes shape actual features, texture actual features and color actual features, wherein: The actual shape characteristics include actual area, actual circularity, and actual aspect ratio; The actual texture features include the actual energy of gray-level co-occurrence matrix, the actual entropy of gray-level co-occurrence matrix, the actual contrast of gray-level co-occurrence matrix and the actual correlation of gray-level co-occurrence matrix; The actual characteristics of color include actual hue, actual saturation, and actual lightness; A shape similarity function is constructed based on the shape features and the actual shape features to determine the shape similarity value; Constructing a texture similarity function based on texture features and actual texture features to determine a texture similarity value; Construct a color similarity function based on the color feature and the actual color feature to determine the color similarity value; The similarity measurement conditions are: the shape similarity value is not greater than the shape similarity threshold, the texture similarity value is not greater than the texture similarity threshold, and the color similarity value is not greater than the color similarity threshold.

7. The method for localizing stained cells for pathological diagnosis according to claim 6, characterized in that: Shape Similarity Function for: ； in, is the area of ​​the dyed region where the seed point s is located, is the actual area of ​​the dyed region where the neighborhood pixel p is located, for The weight factor, is the circularity of the dyed region where the seed point s is located, is the actual circularity of the dyed area where the neighborhood pixel p is located, for The weight factor, is the aspect ratio of the dyed region where the seed point s is located, is the actual aspect ratio of the dyed area where the neighborhood pixel p is located, for The weight factor of Texture similarity function for: ； in, is the gray-level co-occurrence matrix energy of the dyed region where the seed point s is located, is the actual energy of the gray-level co-occurrence matrix of the dyed area where the neighborhood pixel p is located, is the gray-level co-occurrence matrix entropy of the dyed region where the seed point s is located, is the actual entropy of the gray-level co-occurrence matrix of the dyed area where the neighborhood pixel p is located, is the gray level co-occurrence matrix contrast of the dyed area where the seed point s is located, is the actual contrast ratio of the gray-level co-occurrence matrix of the stained area where the neighborhood pixel p is located, is the gray level co-occurrence matrix correlation of the dyed area where the seed point s is located, The actual correlation of the gray-level co-occurrence matrix of the stained area where the neighborhood pixel p is located; Color similarity function for: ； in, are the coordinates of the seed point s in the HSV color space, representing hue, saturation, and brightness, respectively. are the coordinates of the neighborhood pixel point p in the HSV color space, representing the actual hue, actual saturation, and actual brightness respectively.

8. The method for localizing stained cells for pathological diagnosis according to claim 4, characterized in that: The preliminary segmented cell image is optimized based on the watershed algorithm to obtain the segmented pathological staining image, including the following steps: The Sobel operator is used to calculate the horizontal and vertical gradients of the preliminary segmented cell image. For each pixel, the horizontal gradient and vertical gradient , the calculation formula is as follows: ； ； in, The original pathological staining image at pixel The gray value at Represents the convolution operation, calculating the gradient amplitude of each pixel , ; Traverse each cell region in the preliminary segmented cell image and determine the area of ​​each cell region and area circularity ; If the circularity of the region Greater than the set circularity threshold and less than 1, and the area meets the , then the gradient amplitude at the boundary of the cell region is multiplied by the first correction coefficient, and the first correction coefficient is less than 1, where is the minimum area of ​​the set region. The maximum area of ​​the set region; If the circularity of the region Greater than 1, and the area satisfies , then the gradient amplitude at the boundary of the cell region is multiplied by the second correction coefficient, and the second correction coefficient is less than 1; Get the corrected gradient map; The corrected gradient map is subjected to watershed transformation to obtain the segmented pathological staining image.

9. The method for localizing stained cells for pathological diagnosis according to claim 1, characterized in that: Cell localization based on the segmented pathological staining image includes the following steps: Determine the segmented cell area in the segmented pathological staining image, calculate the geometric center coordinates of each segmented cell area, and record them as the cell center coordinates ; ； in, is the coordinate of the ath pixel in the segmented cell region, and n is the total number of pixels in the segmented cell region; The cell center coordinates are marked based on the set marking type.

10. A stained cell positioning system for pathological diagnosis, applied to the stained cell positioning method for pathological diagnosis according to any one of claims 1 to 9, characterized in that: include: A preprocessing module, which acquires the original pathological staining image based on the image acquisition device and performs preprocessing to obtain a processed pathological staining image; The feature extraction module performs feature processing on the processed pathological staining image to obtain the binary image of the staining area and the staining cell feature data of the pixel points. The staining cell feature data includes shape features, texture features and color features. A segmentation module performs image segmentation on the stained area binary image based on the stained cell feature data of the pixel points to obtain a segmented pathological staining image; The positioning module performs cell positioning based on the segmented pathological staining images.

Images