Diabetic retinopathy pericyte apoptosis image detection method and system
By identifying and correcting blank background areas in diabetic retinopathy images and generating differential correction parameters, the problem of image inconsistency under different acquisition conditions is solved, improving the accuracy of pericyte apoptosis detection and the reliability of detection results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANDONG FIRST MEDICAL UNIV & SHANDONG ACADEMY OF MEDICAL SCI
- Filing Date
- 2026-03-27
- Publication Date
- 2026-06-26
Smart Images
Figure CN122289097A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of image processing and medical image analysis technology, and more specifically, to a method and system for detecting pericyte apoptosis in diabetic retinopathy. Background Technology
[0002] Diabetic retinopathy (DR) is a common microvascular complication of diabetes, severely affecting patients' vision and even leading to blindness. In pathological studies of DR, researchers often observe retinal microvascular-related cells through microscopic images and use pericyte apoptosis and its quantity as reference indicators of early changes in DR, while also evaluating the efficacy of different treatment regimens. To improve sample processing speed and research efficiency, the same batch of slides is usually stained and photographed at different times, under different laboratory conditions, or with different equipment.
[0003] Subsequently, the system identifies potential pericyte-related regions in the image, forming a set of candidate regions. It then performs apoptosis discrimination and quantification on these candidate regions, providing results such as apoptosis / non-apoptosis classification, or grading and counting results. However, in practice, high-throughput samples often come from different batches and under different acquisition conditions. For example, some images may have been taken with a bright-field microscope, while others may have been taken with a fluorescence microscope; some images may have been taken in the morning, while others in the afternoon; some images may have used newly changed light sources or different exposure parameters; some sections may have a yellowish or bluish background due to differences in staining time; and the same staining scheme may result in variations in color development depth depending on the operator. These differences can lead to significant variations in brightness distribution, color distribution, background noise, and local contrast in images of the same type of sample. If the system fails to recognize these distribution changes caused by differences in acquisition conditions during the image status evaluation stage and still sends all images to the same set of preprocessing and standardization rules, inconsistent standardization results may occur. For example, after some images are corrected, the pericyte boundaries become lighter, the nucleus region becomes too bright, and the background texture is magnified in some images. This causes a shift in the image representation on which subsequent candidate region localization and apoptosis discrimination depend. The counting and grading results output by the system will be difficult to keep consistent, the workload of review will increase, and unexplained differences will appear when cross-batch controls are used, which seriously affects the reliability of early diagnostic indicator extraction and efficacy evaluation conclusions.
[0004] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this application provides a method and system for detecting pericyte apoptosis in diabetic retinopathy images. This method effectively corrects differences in brightness, color, background noise, and local contrast of images acquired under different conditions, thereby improving the accuracy and consistency of subsequent pericyte apoptosis discrimination.
[0006] In a first aspect, this application provides a method for detecting pericyte apoptosis images in diabetic retinopathy, comprising: Images of diabetic retinopathy to be processed are acquired using a microscopic image acquisition device; Identify non-cell-covered blank background areas in images of diabetic retinopathy; Extract color and grayscale information from blank background areas; The color deviation information of color information relative to the preset standard color information is calculated, and the gray value deviation information of gray value information relative to the preset standard gray value information is calculated. Based on the color deviation information and gray value deviation information, correction parameters for correcting diabetic retinopathy images are generated. Correction of diabetic retinopathy images based on correction parameters.
[0007] The above scheme can effectively correct the differences in brightness, color, background noise and local contrast of images under different acquisition conditions, thereby improving the accuracy and consistency of subsequent pericyte apoptosis discrimination.
[0008] Furthermore, this application also proposes a step for identifying non-cell-covered blank background regions in diabetic retinopathy images, including: Convert images of diabetic retinopathy to grayscale; Perform local texture analysis on the grayscale image to identify pixels with grayscale values below a first preset threshold as flat pixels; Connectivity analysis is performed on flat pixels, and connected regions with an area ratio exceeding a second preset threshold are identified as blank background regions.
[0009] The above method can more accurately identify blank background areas in images, providing a reliable foundation for subsequent color and grayscale information extraction.
[0010] Furthermore, this application also proposes, after identifying the blank background area, to include: Outlier detection is performed on pixels in blank background areas to remove abnormal pixels whose color or grayscale value distribution deviates from the preset range; The steps for extracting color and grayscale information from a blank background area include: Extract color and grayscale information from the blank background area after removing abnormal pixels.
[0011] The above method can remove abnormal pixels in blank background areas, improving the accuracy and representativeness of extracted color and grayscale information.
[0012] Furthermore, this application also proposes that the color information includes average hue, average saturation, and average brightness; The steps for extracting color and grayscale information from a blank background area include: Calculate the average component values of pixels in the red, green and blue channels within the blank background area, and convert the average component values into average hue, average saturation and average brightness. Grayscale information is obtained by calculating the average grayscale value of pixels within the blank background area.
[0013] The above method can comprehensively extract the color and grayscale features of the blank background area, providing richer information for the subsequent generation of correction parameters.
[0014] Furthermore, this application proposes that the correction parameters are differentiated correction values generated for different coordinate positions of the diabetic retinopathy image, so that pixels located at different coordinate positions of the diabetic retinopathy image can obtain corresponding correction amplitudes.
[0015] The above method enables refined correction of image background inhomogeneity, improving the accuracy and adaptability of the correction.
[0016] Furthermore, this application also proposes a step for generating correction parameters for correcting images of diabetic retinopathy based on color deviation information and grayscale value deviation information, including: Based on pixels at different coordinate positions in diabetic retinopathy images, a background feature distribution trend is fitted and generated; Based on the differences between the background feature distribution trend and the color deviation information and grayscale value deviation information, the corresponding correction parameters are calculated for each pixel position.
[0017] The above method can generate more realistic differential correction parameters based on the distribution trend of local background features in the image.
[0018] Furthermore, this application also proposes a step for fitting and generating a background feature distribution trend based on pixels at different coordinate positions in a diabetic retinopathy image, including: Diabetic retinopathy images were divided into multiple grid cells; Within each grid cell, identify local pixel samples belonging to the blank background area and record their corresponding coordinate positions; Based on the coordinate positions of local pixel samples, a background feature distribution trend is fitted and generated.
[0019] The above method can capture the global distribution trend of image background features more accurately by sampling and fitting local pixel samples.
[0020] Furthermore, this application also proposes a step for fitting and generating a background feature distribution trend based on the coordinate positions corresponding to local pixel samples, including: Establish a bivariate function with the coordinate position of the diabetic retinopathy image as the independent variable; The parameters of the binary function are fitted based on the coordinate positions of the local pixel samples to generate a background feature distribution surface as the background feature distribution trend.
[0021] The above scheme can accurately describe the continuous distribution of image background features through mathematical model fitting, providing a reliable basis for differential correction.
[0022] Furthermore, this application also proposes a step for correcting diabetic retinopathy images based on correction parameters, including: Obtain the raw pixel values of each pixel in the red, green, and blue channels of a diabetic retinopathy image; The original pixel value of each pixel is adjusted pixel by pixel according to the correction parameters so that the adjusted pixel value is limited to a preset valid value range, and pixel values that exceed the preset valid value range are truncated.
[0023] The above method ensures that the corrected pixel values are within the effective range, avoiding image distortion caused by over-correction.
[0024] Secondly, this application also provides an image detection system for pericyte apoptosis in diabetic retinopathy, comprising: The image acquisition module is used to acquire images of diabetic retinopathy to be processed through a microscopic image acquisition device; The blank background region recognition module is used to identify non-cell-covered blank background regions in diabetic retinopathy images. The information extraction module is used to extract color information and grayscale value information from blank background areas; The correction parameter generation module is used to calculate the color deviation information of color information relative to the preset standard color information, and to calculate the gray value deviation information of gray value information relative to the preset standard gray value information. Based on the color deviation information and gray value deviation information, correction parameters for correcting diabetic retinopathy images are generated. The image correction module is used to correct images of diabetic retinopathy based on correction parameters.
[0025] The above approach provides a system-level solution for detecting pericytic apoptosis in diabetic retinopathy, improving detection efficiency and accuracy.
[0026] As can be seen from the above, the method and system for detecting pericyte apoptosis in diabetic retinopathy provided in this application acquires the diabetic retinopathy image to be processed through a microscopic image acquisition device, identifies the blank background area not covered by cells, extracts color information and grayscale value information from the blank background area, calculates the color deviation information of the color information relative to the preset standard color information, and calculates the grayscale value deviation information of the grayscale value information relative to the preset standard grayscale value information. Based on the color deviation information and grayscale value deviation information, correction parameters for correcting the diabetic retinopathy image are generated. Based on the correction parameters, the diabetic retinopathy image is corrected, which can effectively correct the differences in brightness, color, background noise and local contrast of the image under different acquisition conditions, and improve the accuracy and consistency of subsequent pericyte apoptosis discrimination. Attached Figure Description
[0027] Figure 1 This is a schematic flowchart of a method for detecting pericyte apoptosis in diabetic retinopathy, provided in an embodiment of this application.
[0028] Figure 2 This is a schematic diagram of the structure of a pericyte apoptosis image detection system for diabetic retinopathy provided in an embodiment of this application.
[0029] Labeling Explanation: 210, Image Acquisition Module; 220, Blank Background Area Recognition Module; 230, Information Extraction Module; 240, Correction Parameter Generation Module; 250, Image Correction Module. Detailed Implementation
[0030] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.
[0031] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0032] In the pathological study of diabetic retinopathy, observing and quantifying the apoptosis of pericytes in retinal microvessels is a crucial step in assessing early disease changes and the effectiveness of treatment. To handle the large volume of samples, researchers typically employ high-throughput processing, aggregating and analyzing microscopic images from different batches, at different time points, and even from different laboratories using different equipment. However, this high-throughput approach presents a significant technical challenge: the inconsistency in image representation.
[0033] Specifically, the image acquisition process is highly susceptible to interference from various environmental factors. For example, when a microscope is used in the morning and afternoon, the color temperature and intensity of its light source may vary slightly due to fluctuations in mains voltage or the decay of the bulb after prolonged use. Images taken in the morning may have a bluish background and higher brightness, while images taken in the afternoon may have a yellowish background and lower brightness. Furthermore, slight differences in the chemical composition of different batches of staining solutions, or deviations in staining time control when performed by different operators, can all lead to different background colors in the final slides, some yellowish and some bluish. In addition, different models of microscopes or digital cameras have different optical systems and photosensitive elements, which further exacerbates the differences in color, contrast, and brightness distribution in the images.
[0034] Problems arise when images from diverse sources and exhibiting varying characteristics are fed into a unified automated analysis workflow. Traditional preprocessing methods typically standardize all images using a fixed set of parameters, such as uniformly stretching brightness or enhancing contrast. This one-size-fits-all approach cannot adapt to the unique acquisition environment of each image. An image with moderate brightness may become overexposed after applying brightness enhancement parameters designed for dim images, resulting in the loss of fine edge contours and internal texture details of pericytes. Conversely, an image that is already dark may still be insufficiently bright after processing, or although brightness is enhanced, inherent color cast problems may remain, making it difficult to accurately identify cell structures against abnormally colored backgrounds.
[0035] This failure of standardization directly leads to a shift in the image features relied upon for subsequent pericyte apoptosis discrimination. Color or brightness thresholds used to identify apoptotic cells may be effective on one image but completely fail on another improperly calibrated image. This causes unexplained fluctuations in apoptosis counts and grading results for the same sample when automated analysis systems process batches, with some apoptotic areas being missed and non-apoptotic areas being misclassified. Researchers are forced to invest significant time in manual verification, which not only reduces efficiency but, more seriously, significantly undermines the reliability of analytical conclusions when longitudinal comparisons of treatment efficacy are needed between different batches of samples.
[0036] Firstly, see [the following] Figure 1 This application provides a method for detecting pericyte apoptosis images in diabetic retinopathy, comprising: S1. Acquire images of diabetic retinopathy to be processed using a microscopic image acquisition device; S2. Identify non-cell-covered blank background areas in diabetic retinopathy images; S3. Extract color and grayscale information from the blank background area; S4. Calculate the color deviation information of the color information relative to the preset standard color information, and calculate the gray value deviation information of the gray value information relative to the preset standard gray value information. Generate correction parameters for correcting diabetic retinopathy images based on the color deviation information and the gray value deviation information. S5. Correct diabetic retinopathy images based on correction parameters.
[0037] Specifically, images of retinal lesion sections to be analyzed are captured using microscopic image acquisition devices such as digital cameras connected to microscopes. To preserve the original pixel information to the greatest extent possible, the acquired images are typically stored in lossless or low-loss compressed formats, such as TIFF or high-quality BMP formats. After image acquisition, the crucial blank background region identification stage begins, accurately locating regions within the complex cellular and tissue structures that contain only background information from the slide and staining solution. These regions are the foundation for subsequent corrections. After identifying these blank background regions as environmental fingerprints, feature information that quantifies their visual representation, namely color and grayscale information, is extracted. This information directly reflects the light source intensity and the inherent color cast of the staining solution during image acquisition. Next, the extracted actual background information is compared with a preset standard state representing ideal acquisition conditions. This standard state can be understood as what a perfect background, captured under standard lighting, without any color cast, should look like. For example, a near-pure white standard color information and a corresponding high-brightness standard grayscale value information can be preset. By comparing actual values with standard values, the deviation of the current image in terms of color and brightness can be accurately calculated. This deviation is the color deviation information and grayscale value deviation information. The preset standard color information and preset standard grayscale value information represent the physicochemical characteristics that a blank background should present under ideal imaging conditions. In practical applications, standard color information is usually defined as a set of color vectors representing a pure light source without color deviation, such as a white balance state with extremely high values in the red, green, and blue channels. Standard grayscale value information corresponds to the brightness benchmark after mapping this color vector. These two preset indicators serve as the calibration endpoint, ensuring that images taken at different times and with different devices, after processing, can all align to this consistent background state, thereby eliminating the problem of inconsistent visual background color caused by differences in ambient light or color batches.
[0038] Based on the calculated deviation information, a set of correction parameters is generated for image correction. These parameters function as inverse compensation; if a yellowish background is detected, the correction parameters are set to reduce the yellow component in the image; if the overall image is detected to be too dark, the correction parameters are set to increase the overall brightness of the image. This ensures that the correction operation is targeted and tailored to each image. Color deviation information reflects the vector difference between the actual acquired background hue and the ideal white balance. For example, when a slice has a yellowish background due to prolonged staining, this deviation information records the excessive superposition of the yellow component in the red and green channels. Grayscale deviation information reflects the deviation in the overall exposure of the image, used to quantify the brightness shift caused by excessive or insufficient lighting. These two types of deviation information are obtained by sampling multiple points in the identified blank background areas and taking the average or median value, forming the original data basis for generating subsequent inverse compensation parameters.
[0039] Finally, this set of personalized correction parameters was applied to the entire image of diabetic retinopathy to be processed. This application process was performed pixel-by-pixel to ensure that every point in the image was properly adjusted. After adjustment, the color and brightness inconsistencies in the original image were effectively eliminated.
[0040] Through the aforementioned series of steps, images from different sources and under different acquisition conditions have had their background distribution, color representation, and brightness range corrected to a highly uniform state. This provides a stable and reliable input for the subsequent automatic pericyte apoptosis discrimination algorithm, avoiding the failure of discrimination criteria due to inconsistent backgrounds. This significantly improves the accuracy and reliability of early diagnosis and efficacy assessment, effectively eliminating analytical errors caused by differences in image acquisition conditions and ensuring the accuracy and comparability of high-throughput detection results.
[0041] Furthermore, the steps for identifying non-cell-covered blank background regions in diabetic retinopathy images include: Convert images of diabetic retinopathy to grayscale; Perform local texture analysis on the grayscale image to identify pixels with grayscale values below a first preset threshold as flat pixels; Connectivity analysis is performed on flat pixels, and connected regions with an area ratio exceeding a second preset threshold are identified as blank background regions.
[0042] Specifically, the original color image is converted to a grayscale image, stripping away the color dimension and focusing the analysis on the brightness and intensity of pixels. This is because the most prominent feature of background regions is the smooth variation in pixel intensity, indicating a lack of texture. Next, local texture analysis is performed on the converted grayscale image. This process is achieved by sliding a computational window across the image. For example, a 5-pixel multi-pixel window can be set, and for each pixel in the image, the standard deviation of the grayscale values of all pixels within its surrounding window is calculated. Standard deviation is a measure of data dispersion; in an image, the smaller the standard deviation of a region, the more uniform the pixel grayscale values are, and the flatter the texture. Therefore, pixels with a local standard deviation below a preset first threshold are initially identified as flat pixels. These flat pixels are candidates for constituting background regions. Local texture analysis aims to quantify the complexity of individual micro-regions in an image using statistical methods. In this process, the fluctuation range of pixel intensity within a sliding neighborhood centered on a specific pixel is calculated to determine whether the region contains cellular structures. The first preset threshold serves as a watershed for determining flatness. When the intensity fluctuation or absolute value of a pixel and its surrounding area falls within the range defined by this threshold, it indicates that the location lacks obvious tissue texture or edge features. This judgment logic can effectively eliminate morphologically characteristic areas from complex images filled with retinal microvessels and nerve fibers, focusing the recognition on the pure slide background.
[0043] However, simply identifying flat pixels is insufficient, as tiny, flat gaps may exist between cells. To eliminate these fragmented, non-background regions, connected component analysis is required. This analysis connects all adjacent flat pixels into individual regions. The area of each connected region is then calculated. Only those connected regions with sufficiently large areas are considered true blank background regions. This size criterion is defined by a second preset threshold; for example, only connected regions with an area exceeding 10% of the total number of pixels in the image can be ultimately identified as non-cell-covered blank background regions. In this way, small noise regions can be effectively filtered out, ensuring that the selected background regions are sufficiently representative.
[0044] Furthermore, after identifying the blank background area, the process includes: Outlier detection is performed on pixels in blank background areas to remove abnormal pixels whose color or grayscale value distribution deviates from the preset range; The steps for extracting color and grayscale information from a blank background area include extracting color and grayscale information from the blank background area after removing abnormal pixels.
[0045] In actual slide preparation and imaging processes, even large blank background areas can be contaminated by microscopic impurities. For example, undissolved pigment precipitates may exist in the staining solution, and tiny air bubbles or fibrous dust may be introduced during mounting. The color and brightness of these impurities usually differ significantly from the true background. Including them in the extracted background information sample is like a drop of ink contaminating a glass of water, causing a shift in the calculated average color and brightness values, and consequently generating incorrect correction parameters.
[0046] To address this, before formally extracting color and grayscale information, this application first performs outlier detection on all pixels within the identified blank background area. This detection process analyzes the distribution patterns of pixel groups within the area in terms of color and grayscale values. If the color or grayscale value of a pixel significantly deviates from the concentrated distribution range of most pixels—for example, a dark dye particle might have a grayscale value much lower than the background average, while a bright bubble reflection might have a grayscale value much higher than the background average—then these pixels will be identified as anomalous pixels or outliers.
[0047] Once these anomalous pixels are identified, they are removed from the background sample set. Then, color and grayscale information are extracted from the purified background area from which anomalous pixels have been removed. This process ensures that the extracted background features accurately reflect the microscope's optical path environment and the staining background color, without being interfered with by randomly occurring microscopic impurities, thereby greatly improving the stability and accuracy of the calibration benchmark.
[0048] Furthermore, color information includes average hue, average saturation, and average brightness; The steps for extracting color and grayscale information from a blank background area include: Calculate the average component values of pixels in the red, green and blue channels within the blank background area, and convert the average component values into average hue, average saturation and average brightness. Grayscale information is obtained by calculating the average grayscale value of pixels within the blank background area.
[0049] To comprehensively and accurately describe the visual characteristics of the background, this application specifically defines the composition and extraction methods of color information and grayscale value information. Color information is defined as three components: average hue, average saturation, and average brightness. These three components together constitute a complete description of color attributes. Hue represents the basic category of the color, such as whether it leans towards red or blue; saturation represents the purity or vividness of the color; and brightness represents the lightness or darkness of the color.
[0050] First, for all pixels within the purified blank background area, calculate the arithmetic mean of their pixel values in the red, green, and blue channels, respectively, to obtain an average red component value, an average green component value, and an average blue component value. These three averages constitute the average RGB color of the background area. Then, using a standard color space conversion algorithm, convert this average RGB value into its corresponding values in the HSB color space, which better matches human visual perception: average hue, average saturation, and average brightness.
[0051] At the same time, to quantify the overall brightness of the background, grayscale information also needs to be extracted. This can be obtained by calculating the average grayscale value of all pixels within the blank background area. The calculation method can be to first convert the RGB value of each pixel to its corresponding grayscale value, and then calculate the average of all grayscale values; or directly use the already calculated average RGB component values, and apply a weighted average formula, for example, grayscale value equals 0.299 multiplied by the average red component value plus 0.587 multiplied by the average green component value plus 0.114 multiplied by the average blue component value, to obtain the final grayscale information.
[0052] Furthermore, the correction parameters are differentiated correction values generated for different coordinate positions in the diabetic retinopathy image, so that pixels located at different coordinate positions in the diabetic retinopathy image receive corresponding correction magnitudes.
[0053] The aforementioned correction methods can effectively address overall and global image differences caused by different batches and devices. However, in some cases, background inconsistencies are not only present between images but may also exist within a single image. A typical example is the edge vignetting caused by the microscope's optical system, also known as halos. This phenomenon causes the central area of the image to be brighter than the surrounding edge areas. If a globally uniform correction parameter is still used to adjust the entire image in this case, a dilemma arises: if the correction is based on the central area, the edge areas will still be insufficiently bright; if the edge areas are used as the reference, the central area will become overexposed. This local deviation can also interfere with subsequent cell identification.
[0054] To address this issue of inhomogeneity within images, this application does not generate a single correction parameter applicable to the entire image. Instead, it generates differentiated correction values for different coordinate locations within the image. This means that pixels located at different positions in the image will receive a customized correction magnitude based on their local environment. For example, in an image with edge vignetting, pixels in dark corners will receive a larger brightness boost, while pixels in bright centers may receive only a small adjustment, or even no adjustment at all. In this way, precise smoothing of background unevenness within the image can be achieved.
[0055] Furthermore, the step of generating correction parameters for correcting diabetic retinopathy images based on color deviation information and grayscale value deviation information includes: Based on pixels at different coordinate positions in diabetic retinopathy images, a background feature distribution trend is fitted and generated; Based on the differences between the background feature distribution trend and the color deviation information and grayscale value deviation information, the corresponding correction parameters are calculated for each pixel position.
[0056] This application no longer treats the background as a uniform whole, but rather as a continuously varying surface in two-dimensional space. First, based on background pixel samples at different coordinate positions in the image, a model describing the distribution trend of the entire image's background features is generated through mathematical fitting. This model can be a mathematical function or a data surface, capable of predicting the background brightness or color value at any coordinate point in the image. For example, for edge vignetting, the fitted background brightness distribution trend model might appear as a dome-shaped surface, high in the center and low around the edges. The background feature distribution trend is a spatial model of the entire image's background representation, used to describe the evolution of brightness or color at different locations in the image. A binary function is the mathematical tool for this modeling, using the pixel's horizontal and vertical coordinates as independent variables and the background's feature attributes as dependent variables. By constructing a second-order or higher-order surface function and substituting background sample points at known locations for regression calculations, a feature surface covering the entire image can be fitted. This surface can not only reflect global deviations, but also accurately capture edge glare phenomena caused by poor optical lens quality, such as a bright center and dark edges, providing a basis for subsequent fine position compensation.
[0057] After obtaining this background feature distribution trend model, a specific correction parameter can be calculated for each pixel location in the image. The calculation is based on the difference between the background feature value predicted by the model at that location and the preset standard background feature value. In other words, for each pixel, the local background state at its location is compared with the ideal state, and the difference that needs to be compensated is calculated. This difference is the correction parameter corresponding to that pixel location. In this way, it can be ensured that the correction operation can accurately adapt to local changes within the image, achieving complete elimination of background non-uniformity.
[0058] Furthermore, the step of fitting and generating the background feature distribution trend based on pixels at different coordinate positions in the diabetic retinopathy image includes: Diabetic retinopathy images were divided into multiple grid cells; Within each grid cell, identify local pixel samples belonging to the blank background area and record their corresponding coordinate positions; Based on the coordinate positions of local pixel samples, a background feature distribution trend is fitted and generated.
[0059] First, the entire image of diabetic retinopathy to be processed is logically divided into multiple grid units, forming a chessboard-like structure, for example, divided into 10 rows and 10 columns, for a total of 100 grids. The purpose of this division is to decompose the global analysis task into a series of local analysis tasks.
[0060] Next, within each grid cell, the aforementioned blank background region identification process is executed independently to identify local pixel samples that belong to the background within that grid. A crucial step in identifying these local samples is recording their coordinate positions within the entire image. This yields a series of background feature data points with spatial location labels.
[0061] Finally, using all these collected local pixel samples with coordinate information, mathematical fitting is performed to generate a background feature distribution trend covering the entire image. Since the sample points are distributed across various regions of the image, the fitted model can effectively capture the gradual change of background features from one region to another, providing a solid data foundation for subsequent fine-tuning.
[0062] Furthermore, the step of fitting and generating the background feature distribution trend based on the coordinate positions corresponding to local pixel samples includes: Establish a bivariate function with the coordinate position of the diabetic retinopathy image as the independent variable; The parameters of the binary function are fitted based on the coordinate positions of the local pixel samples to generate a background feature distribution surface as the background feature distribution trend.
[0063] Specifically, a bivariate function is first established with the coordinates of the image as the independent variable. The form of this function can be chosen according to actual needs; for example, to simulate a centrally symmetric distribution such as edge vignetting, a second-order bivariate polynomial function can be used. Then, the coordinates of all local pixel samples collected in the previous step and their corresponding background feature values are used as known data points and substituted into this bivariate function model. Regression analysis algorithms such as least squares are used to fit the function parameters, calculating a set of parameter values that minimizes the overall error between the function curve and all data points. Once this optimal set of parameters is determined, the bivariate function is fully defined. At this point, the mathematical surface described by the function is a smooth, continuous background feature distribution surface, which can serve as the final background feature distribution trend. Using this function, the background feature value at any pixel location in the image can be calculated, providing a precise local reference for pixel-by-pixel differential correction.
[0064] Furthermore, the steps for correcting diabetic retinopathy images based on correction parameters include: Obtain the raw pixel values of each pixel in the red, green, and blue channels of a diabetic retinopathy image; The original pixel value of each pixel is adjusted pixel by pixel according to the correction parameters so that the adjusted pixel value is limited to a preset valid value range, and pixel values that exceed the preset valid value range are truncated.
[0065] After generating the correction parameters for each pixel, the final step is to apply them to the original image to complete the correction process. This application process is a fine-grained, pixel-by-pixel operation.
[0066] First, iterate through each pixel in the image and obtain its original pixel values in the red, green, and blue channels. These three values constitute the original color of the pixel. Then, based on the pixel's coordinates, find or calculate a custom correction parameter for it. Apply this correction parameter to the original pixel value. For example, if the correction parameter is a value representing a brightness increase, add this value to the pixel's three color channel values; if the correction parameter is a vector for color balance, perform the corresponding mathematical operations between this vector and the pixel's RGB vector.
[0067] During the adjustment process, it's important to note that digital image pixel values typically have a fixed valid range. For example, for an 8-bit image, the value of each color channel must be between 0 and 255. Correction operations may cause the calculation results to exceed this range. For instance, adding a positive correction value to an already bright pixel value might result in a value greater than 255. To ensure the corrected image data is valid, pixel values exceeding the preset valid range need to be truncated. Specifically, any calculated value less than 0 is forcibly set to 0, and any value greater than 255 is forcibly set to 255. After this processing, the final corrected image will have all pixel values within the valid range, allowing for normal display and further analysis. Truncating the correction parameters is crucial when applying correction parameters to the original pixel values during addition or subtraction operations. Since the calculation results may exceed the numerical range that the digital image can represent, truncation must be performed. For common 8-bit depth images, the valid range of pixel values is limited to 0 to 255. If the corrected value is below zero, it is forcibly assigned a value of zero to prevent artifacts caused by reverse overflow; if the value exceeds 255, it is forcibly assigned a value of 255 to prevent highlight color distortion. This process ensures that the corrected image maintains the integrity of its data structure and can be correctly read and parsed by subsequent pericyte apoptosis discrimination algorithms.
[0068] Secondly, see Figure 2 This application also provides an image detection system for pericyte apoptosis in diabetic retinopathy, comprising: Image acquisition module 210 is used to acquire images of diabetic retinopathy to be processed through a microscopic image acquisition device; Blank background region recognition module 220 is used to identify non-cell-covered blank background regions in diabetic retinopathy images; Information extraction module 230 is used to extract color information and grayscale value information from the blank background area; The correction parameter generation module 240 is used to calculate the color deviation information of color information relative to the preset standard color information, and to calculate the gray value deviation information of gray value information relative to the preset standard gray value information, and to generate correction parameters for correcting diabetic retinopathy images based on the color deviation information and the gray value deviation information. Image correction module 250 is used to correct images of diabetic retinopathy based on correction parameters.
[0069] This application accurately identifies blank background areas, quantifies their color and brightness characteristics, and compares them with a preset ideal standard state to calculate the specific deviations in the current image. Subsequently, a set of inverse compensation parameters specifically designed to offset these deviations is generated, and the entire image undergoes fine-tuning at the pixel level. After this processing, regardless of whether the original image is yellowish or bluish, too bright or too dark, its background will be corrected to a uniform, neutral standard state. This provides a stable and consistent visual foundation for subsequent pericyte apoptosis discrimination algorithms, fundamentally eliminating analytical errors caused by differences in acquisition conditions and ensuring the accuracy and comparability of high-throughput detection results.
[0070] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.
Claims
1. A method for detecting pericyte apoptosis in diabetic retinopathy images, characterized in that, include: Images of diabetic retinopathy to be processed are acquired using a microscopic image acquisition device; Non-cell-covered blank background areas were identified in the diabetic retinopathy images; Extract color information and grayscale value information from the blank background area; Calculate the color deviation information of the color information relative to the preset standard color information, and calculate the gray value deviation information of the gray value information relative to the preset standard gray value information. Generate correction parameters for correcting the diabetic retinopathy image based on the color deviation information and the gray value deviation information. The diabetic retinopathy image is corrected based on the correction parameters.
2. The method for detecting pericyte apoptosis in diabetic retinopathy according to claim 1, characterized in that, The step of identifying non-cell-covered blank background regions in the diabetic retinopathy image includes: Convert the diabetic retinopathy image to a grayscale image; Perform local texture analysis on the grayscale image and identify pixels in the grayscale image whose grayscale value is lower than a first preset threshold as flat pixels; Connectivity analysis is performed on the flat pixels, and connected components with an area ratio exceeding a second preset threshold are identified as the blank background region.
3. The method for detecting pericyte apoptosis in diabetic retinopathy according to claim 1, characterized in that, After identifying the blank background area, the process includes: Outlier detection is performed on the pixels in the blank background area to remove abnormal pixels whose color or grayscale value distribution deviates from a preset range; The step of extracting color information and grayscale value information from the blank background area includes: Extract the color information and grayscale information from the blank background area after removing the abnormal pixels.
4. The method for detecting pericyte apoptosis in diabetic retinopathy according to claim 1, characterized in that, The color information includes average hue, average saturation, and average brightness. The step of extracting color information and grayscale value information from the blank background area includes: Calculate the average component values of pixels in the red, green, and blue channels within the blank background area, and convert the average component values into the average hue, average saturation, and average brightness. The grayscale information is obtained by calculating the average grayscale value of the pixels within the blank background area.
5. The method for detecting pericyte apoptosis in diabetic retinopathy according to claim 1, characterized in that, The correction parameters are differential correction values generated for different coordinate positions of the diabetic retinopathy image, so that pixels located at different coordinate positions of the diabetic retinopathy image obtain corresponding correction amplitudes.
6. The method for detecting pericyte apoptosis in diabetic retinopathy according to claim 5, characterized in that, The step of generating correction parameters for correcting the diabetic retinopathy image based on the color deviation information and the grayscale deviation information includes: Based on the pixels at different coordinate positions in the diabetic retinopathy image, a background feature distribution trend is fitted and generated; Based on the differences between the background feature distribution trend and the color deviation information and the grayscale value deviation information, the corresponding correction parameter is calculated for each pixel position.
7. The method for detecting pericyte apoptosis in diabetic retinopathy according to claim 6, characterized in that, The step of fitting and generating a background feature distribution trend based on pixels at different coordinate positions in the diabetic retinopathy image includes: The diabetic retinopathy image is divided into multiple grid cells; Within each grid cell, local pixel samples belonging to the blank background region are identified, and their corresponding coordinate positions are recorded; The background feature distribution trend is generated by fitting the coordinate positions corresponding to the local pixel samples.
8. The method for detecting pericyte apoptosis in diabetic retinopathy according to claim 7, characterized in that, The step of fitting and generating the background feature distribution trend based on the coordinate positions corresponding to the local pixel samples includes: Establish a bivariate function with the coordinate position of the image of diabetic retinopathy as the independent variable; The binary function is fitted with parameters based on the coordinate positions corresponding to the local pixel samples to generate a background feature distribution surface as the background feature distribution trend.
9. The method for detecting pericyte apoptosis in diabetic retinopathy according to claim 6, characterized in that, The step of correcting the diabetic retinopathy image based on the correction parameters includes: Obtain the original pixel values of each pixel in the red, green, and blue channels of the diabetic retinopathy image; The original pixel value of each pixel is adjusted pixel by pixel according to the correction parameters so as to limit the adjusted pixel value to a preset effective value range, and the pixel value that exceeds the preset effective value range is truncated.
10. A system for detecting pericyte apoptosis in diabetic retinopathy, characterized in that, include: The image acquisition module is used to acquire images of diabetic retinopathy to be processed through a microscopic image acquisition device; The blank background region recognition module is used to identify non-cell-covered blank background regions in the diabetic retinopathy image. The information extraction module is used to extract color information and grayscale value information from the blank background area; The correction parameter generation module is used to calculate the color deviation information of the color information relative to the preset standard color information, and to calculate the gray value deviation information of the gray value information relative to the preset standard gray value information, and to generate correction parameters for correcting the diabetic retinopathy image based on the color deviation information and the gray value deviation information. An image correction module is used to correct the diabetic retinopathy image based on the correction parameters.