A background noise suppression system for microarray protein chips

Abstract

The application provides a background noise suppression system for a microarray protein chip, and relates to the technical field of chip detection, and comprises a collection module, which is used for collecting an original fluorescence image of the microarray protein chip by a laser confocal scanner at a preset excitation power, and taking the original fluorescence image as a to-be-processed signal matrix; an extraction module, which is used for locating three reference entities pre-solidified on a chip substrate from the to-be-processed signal matrix, the three reference entities being located at non-array regions of the upper left corner, the center point and the lower right corner of the chip respectively, and each reference entity being composed of inert fluorescent microspheres with a known concentration; and the mean values of the fluorescence intensities of local windows where the three reference entities are located are extracted respectively as discrete sampling values of spatial heterogeneity background noise. The application improves the low-abundance protein signal detection rate, reduces the quantitative deviation of the array edge, and improves the detection repeatability and quantitative accuracy.

Description

Technical Field

[0001] This invention relates to the field of biochip detection technology, and in particular to a background noise suppression system for microarray protein chips. Background Technology

[0002] Microarray protein chips are widely used in biomarker screening and high-throughput immunoassay. However, their fluorescence signal acquisition is easily affected by spatially heterogeneous background noise such as substrate autofluorescence, non-uniform scanning light field, and edge light scattering. Current mainstream processing methods mostly adopt global mean subtraction or neighborhood local fitting. Taking the detection of a protein chip for a specific allergen biomarker as an example, after the chip acquires fluorescence images using laser confocal scanning, it directly uses fixed threshold background subtraction and conventional Gaussian filtering for noise reduction. However, this technology has drawbacks: it is difficult to accurately characterize the non-linear background fluctuations on the chip surface from the center to the corners, and it cannot adaptively suppress residual scattering noise at the edges of the array points. This can cause low-abundance protein signals to be masked by noise, and the quantitative deviation at the array edges to be large. Furthermore, global processing can easily destroy the spatial continuity of the effective signal, making it difficult to meet the requirements of precise clinical detection in terms of detection repeatability and quantitative accuracy. Summary of the Invention

[0003] This invention provides a background noise suppression system for microarray protein chips, which improves the detection rate of low-abundance protein signals, reduces quantitative deviation at array edges, and enhances detection repeatability and quantitative accuracy.

[0004] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0005] The first aspect concerns a background noise suppression system for microarray protein chips, including:

[0006] The acquisition module is used to acquire the raw fluorescence image of the microarray protein chip using a laser confocal scanner at a preset excitation power, and to use the raw fluorescence image as a signal matrix to be processed.

[0007] The extraction module is used to locate three pre-cured reference entities on the chip substrate from the signal matrix to be processed. The three reference entities are located in the non-array regions of the upper left corner, center point and lower right corner of the chip, respectively. Each reference entity is composed of inert fluorescent microspheres of known concentration. The average fluorescence intensity of the local window where the three reference entities are located is extracted as discrete sampling values ​​of spatial heterogeneous background noise.

[0008] The module is used to construct a virtual freeform surface with a continuous distribution of background noise on the entire chip surface using a bicubic interpolation algorithm, with the two-dimensional coordinates of three reference entities and the corresponding mean fluorescence intensity as control nodes.

[0009] The partitioning module is used to perform isotherm partitioning on the virtual freeform surface under Riemannian metric, dividing the surface into multiple annular and radial regions along the principal curvature direction; calculating the local gradient tensor of background noise within the annular and radial regions to obtain the normal offset correction coefficient of each pixel position relative to the virtual freeform surface;

[0010] The calculation module is used to perform point-by-point difference operations between the normal offset correction coefficient and the gray value of each corresponding pixel in the original fluorescence image to obtain a preliminary background suppression image;

[0011] The processing module is used to perform anisotropic diffusion filtering on the residual light scattering noise at the edges of each array point in the preliminary background suppression image, and finally outputs a pure fluorescence signal image with suppressed background noise.

[0012] Furthermore, the raw fluorescence images of the microarray protein chip are acquired using a laser confocal scanner at a preset excitation power, and these raw fluorescence images are used as a signal matrix to be processed, including:

[0013] Based on the excitation spectral characteristics of the fluorescent probes labeled on the microarray protein chip, the excitation wavelength of the laser confocal scanner and the aperture of the detection pin are calibrated to match the parameters, and the laser excitation power is locked to the preset excitation power;

[0014] The laser confocal scanner is controlled to perform row-by-row grid scanning on the microarray protein chip, and the photoelectric conversion signal of each scanning point is collected synchronously. Dynamic time-series compensation is performed on the photo-induced attenuation effect during the scanning process to obtain the original pixel dataset.

[0015] The original pixel dataset is mapped to a two-dimensional pixel grid according to the physical scan coordinates. After performing dark current subtraction and photoelectric sensor background noise calibration, it is converted into a signal matrix to be processed, represented by grayscale values.

[0016] Furthermore, three pre-cured reference entities on the chip substrate are located from the signal matrix to be processed. These three reference entities are located in the non-array regions at the upper left, center, and lower right corners of the chip, respectively. Each reference entity is composed of inert fluorescent microspheres of known concentration. The average fluorescence intensity of the local windows containing each of the three reference entities is extracted as discrete sampling values ​​for spatially heterogeneous background noise, including:

[0017] Based on the physical layout coordinate system of the microarray protein chip, the geometric centers of three preset non-array regions—the upper left corner, the center point, and the lower right corner—are retrieved and locked in the signal matrix to be processed, and three reference entities formed by the solidification of inert fluorescent microspheres of known concentration are identified.

[0018] Using the geometric center of each reference entity as the origin, a rectangular local window of a preset size is extracted from the signal matrix to be processed. The boundary of the local window is strictly limited to the non-array area and maintains a preset safe distance from the edge of the microarray.

[0019] Iterate through all pixels covered in each local window, extract the corresponding grayscale values ​​and remove outliers that exceed the preset dynamic threshold range, and calculate the arithmetic mean of the grayscale values ​​of the remaining valid pixels.

[0020] The arithmetic mean of the gray values ​​corresponding to the three local windows is bound to the two-dimensional spatial coordinates of each reference entity to obtain discrete sampled values ​​that characterize the intensity of the spatial heterogeneous background noise distribution on the chip surface.

[0021] Furthermore, using the two-dimensional coordinates of three reference entities and their corresponding mean fluorescence intensities as control nodes, a virtual freeform surface with a continuous distribution of background noise across the entire chip surface is constructed using a bicubic interpolation algorithm, including:

[0022] The discrete sampled values ​​and their bound two-dimensional spatial coordinates are spatially aligned and normalized to construct a control node dataset containing three-dimensional attributes: horizontal physical coordinates, vertical physical coordinates, and mean fluorescence intensity.

[0023] Based on the physical size of the microarray protein chip and the pixel resolution of the original fluorescence image, a two-dimensional regular interpolation grid covering the entire chip surface is established, and the control node dataset is accurately mapped to the corresponding reference grid points in the two-dimensional regular interpolation grid.

[0024] Using each reference grid point as the interpolation center, a cubic polynomial basis function is constructed in the neighborhood through a bicubic interpolation algorithm. By forcing the continuity of function values ​​at matching nodes and the smooth transition condition between first-order and second-order partial derivatives, the fluorescence intensity fitting values ​​of all unfilled grid points in the two-dimensional regular interpolation grid are iteratively calculated.

[0025] The fluorescence intensity fitting values ​​are reconstructed in three dimensions along the axis of the two-dimensional regular interpolation grid to obtain a virtual freeform surface with background noise that covers the entire chip surface and has continuous and smooth curvature.

[0026] Furthermore, isotherm division is performed on the virtual freeform surface using Riemannian metrics, segmenting it into multiple annular and radial regions along the principal curvature directions. Within these annular and radial regions, the local gradient tensor of the background noise is calculated, yielding the normal offset correction coefficient for each pixel position relative to the virtual freeform surface, including:

[0027] By introducing a virtual freeform surface into the Riemann metric space, the parameters of the first and second fundamental forms at each grid point on the surface are calculated to obtain the geometric curvature feature field that characterizes the local undulation of the surface.

[0028] Based on the geometric curvature characteristic field, the principal direction field of the virtual freeform surface is solved. The isotherm topology is divided along the trajectory of the extreme value of the principal curvature, dividing the virtual freeform surface into several annular regions with the same curvature gradient and radial regions radiating along the reference center.

[0029] Within each annular and radial region, the fluorescence intensity distribution of the virtual freeform surface is used as a scalar field. The combination of partial derivatives of the intensity value in the horizontal and vertical coordinates is calculated pixel by pixel to construct a local gradient tensor matrix that characterizes the spatial rate of change and directional features of background noise.

[0030] The local gradient tensor matrix and the unit normal vector of the virtual freeform surface at each pixel coordinate are subjected to vector projection operation. The intensity deviation of each pixel along the normal direction of the surface is extracted. After linear scaling, the normal offset correction coefficient of each pixel position relative to the virtual freeform surface is obtained.

[0031] Furthermore, within each annular and radial region, using the fluorescence intensity distribution of the virtual freeform surface as a scalar field, the partial derivatives of the intensity values ​​in the horizontal and vertical coordinates are calculated pixel-by-pixel to construct a local gradient tensor matrix characterizing the spatial rate of change and directional features of background noise, including:

[0032] Traverse the target pixels in each annular and radial region, and based on the grid topology of the virtual freeform surface, read the horizontal position index, vertical position index and corresponding fluorescence intensity fitting value of each target pixel in the two-dimensional pixel coordinate system.

[0033] Using a preset discrete difference operator template, the center difference value of each target pixel in the horizontal and vertical coordinate axes is calculated respectively, and the horizontal and vertical partial derivative components that characterize the rate of change of background noise intensity along the horizontal and vertical axes are extracted.

[0034] The horizontal and vertical partial derivative components corresponding to the same target pixel are orthogonally combined to construct a two-dimensional gradient vector containing information on the change magnitude and spatial gradient direction.

[0035] The two-dimensional gradient vectors corresponding to all target pixels in each annular region and radial region are matrix-encapsulated in the order of their original physical coordinates to obtain a local gradient tensor matrix covering the entire target region.

[0036] Furthermore, the normal offset correction coefficient is subjected to point-by-point difference operation with the gray values ​​of corresponding pixels in the original fluorescence image to obtain a preliminary background suppression image, including:

[0037] The normal offset correction coefficient is received. Based on the physical coordinate system mapping relationship between the signal matrix to be processed and the virtual freeform surface mesh, the coordinates of each node of the normal offset correction coefficient matrix are spatially registered and aligned with the pixel coordinates of the signal matrix to be processed, and a one-to-one pixel-level mapping index is established.

[0038] Based on pixel-level mapping index, the aligned pixel index positions are traversed, and the initial gray value of the corresponding pixel in the original fluorescence image and the normal offset correction coefficient matching the position are read synchronously to complete the data alignment before the difference operation.

[0039] The normal offset correction coefficient and the corresponding initial gray value are subjected to point-by-point algebraic difference operation to remove the intensity superposition redundancy caused by spatial heterogeneous background noise, and a denoised intermediate gray value dataset carrying effective signal features is obtained.

[0040] The negative gray-level outliers below the zero threshold generated by the difference operation in the denoised intermediate gray-level dataset are truncated and set to zero. The processed effective gray-level values ​​are then rearranged and encapsulated according to the original two-dimensional pixel grid to obtain a preliminary background suppression image.

[0041] Furthermore, anisotropic diffusion filtering is applied to the residual light scattering noise at the edges of each array point in the preliminary background suppression image, ultimately outputting a clean fluorescence signal image with suppressed background noise, including:

[0042] Based on the virtual freeform surface, the Gaussian curvature values ​​of each grid node are extracted, and the Gaussian curvature values ​​are mapped to the pixel array of the signal matrix to be processed according to the physical coordinates to obtain the Gaussian curvature distribution parameter map that characterizes the spatial fluctuation law of background noise.

[0043] Read the preliminary background suppression image, identify the physical boundaries of each array point through the edge contour extraction operator, extract the transition zone pixel region of a preset width outside the boundary, and construct a local noise mask that characterizes the spatial distribution of light scattering residual noise.

[0044] The pixel region covered by the local noise mask is used as the diffusion iteration field. The anisotropic diffusion filtering algorithm is called to perform multiple rounds of pixel gray-level redistribution. In each iteration, the data at the corresponding position of the Gaussian curvature distribution parameter map is dynamically mapped to the adaptive regularization parameter of the diffusion control equation. The diffusion intensity and edge protection threshold are adjusted in real time according to the curvature gradient difference.

[0045] When the diffusion iteration reaches the preset convergence condition or the pixel grayscale residual of adjacent iterations is lower than the preset threshold, the filtering operation is terminated, and the effective grayscale values ​​after iteration convergence are re-encapsulated according to the original two-dimensional pixel grid topology to obtain a pure fluorescence signal image with suppressed background noise.

[0046] In a second aspect, a computing device includes:

[0047] One or more processors;

[0048] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to execute the system.

[0049] Thirdly, a computer-readable storage medium storing a program that, when executed by a processor, performs the system.

[0050] The above-described solution of the present invention has at least the following beneficial effects:

[0051] This method employs three-point reference sampling of the non-array region of the chip, bicubic interpolation to construct a virtual freeform surface for the background noise of the entire chip, and isotherm partitioning under Riemannian measurement to calculate pixel-level normal offset correction coefficients and perform point-by-point differential denoising. Combined with anisotropic diffusion filtering to specifically suppress residual noise from light scattering at the array point edges, it overcomes the technical problems of existing technologies, such as difficulty in accurately characterizing the nonlinear background undulations of the chip surface from the center to the corners, inability to adaptively suppress residual noise from array edge scattering, easy masking of low-abundance protein signals by noise, large quantitative deviation at the array edge, and global processing disrupting the spatial continuity of the effective signal. As a result, it achieves precise suppression of spatially heterogeneous background noise in microarray protein chips, effectively preserves the integrity of the effective signal, improves the detection rate of low-abundance protein signals, reduces quantitative deviation at the array edge, improves detection repeatability and quantitative accuracy, and meets the requirements of precise clinical detection. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of a background noise suppression system for microarray protein chips provided by an embodiment of the present invention.

[0053] Figure 2 This is a schematic diagram of a background noise suppression system for microarray protein chips provided by an embodiment of the present invention. The system performs spatial point-by-point calibration of the corrected echo signal using a two-dimensional spatial attenuation compensation matrix and extracts weak feature signals reflecting human sleep state from the calibrated echo signal.

[0054] Figure 3 This is a simulation diagram of the chip reference entity layout.

[0055] Figure 4 This is a schematic diagram of the background noise sampling statistics of three benchmark entities.

[0056] Figure 5 This is a schematic diagram of the background noise gradient tensor statistics for each region. Detailed Implementation

[0057] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art.

[0058] like Figure 1 As shown, embodiments of the present invention propose a background noise suppression system for microarray protein chips, comprising:

[0059] The acquisition module is used to acquire the raw fluorescence image of the microarray protein chip using a laser confocal scanner at a preset excitation power, and to use the raw fluorescence image as a signal matrix to be processed.

[0060] The extraction module is used to locate three pre-cured reference entities on the chip substrate from the signal matrix to be processed. The three reference entities are located in the non-array regions of the upper left corner, center point and lower right corner of the chip, respectively. Each reference entity is composed of inert fluorescent microspheres of known concentration. The average fluorescence intensity of the local window where the three reference entities are located is extracted as discrete sampling values ​​of spatial heterogeneous background noise.

[0061] The module is used to construct a virtual freeform surface with a continuous distribution of background noise on the entire chip surface using a bicubic interpolation algorithm, with the two-dimensional coordinates of three reference entities and the corresponding mean fluorescence intensity as control nodes.

[0062] The partitioning module is used to perform isotherm partitioning on the virtual freeform surface under Riemannian metric, dividing the surface into multiple annular and radial regions along the principal curvature direction; calculating the local gradient tensor of background noise within the annular and radial regions to obtain the normal offset correction coefficient of each pixel position relative to the virtual freeform surface;

[0063] The calculation module is used to perform point-by-point difference operations between the normal offset correction coefficient and the gray value of each corresponding pixel in the original fluorescence image to obtain a preliminary background suppression image;

[0064] The processing module is used to perform anisotropic diffusion filtering on the residual light scattering noise at the edges of each array point in the preliminary background suppression image, and finally outputs a pure fluorescence signal image with suppressed background noise.

[0065] In this embodiment of the invention, the following technical methods are employed: acquiring original fluorescence images using a laser confocal scanner to construct a signal matrix to be processed; locating three reference points in the non-array region of the chip and extracting the mean fluorescence intensity as a discrete sample value for background noise; constructing a virtual freeform surface for background noise across the entire chip using a bicubic interpolation algorithm; calculating the local gradient tensor and pixel-level normal offset correction coefficient based on Riemannian isotherms; differentially differentiating the correction coefficients from the grayscale values ​​of the original image point by point; and suppressing residual noise from light scattering at the array point edges using anisotropic diffusion filtering. These methods overcome the technical problems in microarray protein chip fluorescence detection, such as the difficulty in accurately characterizing spatially heterogeneous background noise, nonlinear background fluctuation fitting distortion, the inability to adaptively suppress residual noise from array edge scattering, the easy masking of low-abundance protein signals, the destruction of effective signal continuity by global processing, and large quantitative deviations at array edges. This achieves precise suppression of background noise in microarray protein chips, effectively preserving the integrity of the effective signal, improving the detection rate of low-abundance protein signals, reducing quantitative deviations, and improving detection repeatability and quantitative accuracy, thus meeting the requirements for precise detection in biochips.

[0066] In a preferred embodiment of the present invention, a raw fluorescence image of the microarray protein chip is acquired using a laser confocal scanner at a preset excitation power, and the raw fluorescence image is used as a signal matrix to be processed, including:

[0067] Based on the excitation spectral characteristics of the fluorescent probes labeled on the microarray protein chip, the excitation wavelength of the laser confocal scanner and the aperture of the detection pin are calibrated to match the parameters, and the laser excitation power is locked to the preset excitation power. Specifically, this includes: accurately matching and calibrating the core acquisition parameters of the laser confocal scanner according to the excitation spectral characteristics of the fluorescent probes labeled on the microarray protein chip to determine the excitation wavelength that is perfectly matched to the excitation peak of the fluorescent probe; at the same time, adjusting the aperture of the detection pin to a fixed parameter that matches the chip scanning resolution and fluorescence signal focusing requirements to eliminate fluorescence signal acquisition deviation and the introduction of additional background noise caused by parameter mismatch; and stabilizing the laser excitation power to the preset excitation power to avoid abnormal changes in fluorescence signal intensity caused by excitation power fluctuations during scanning, thus avoiding noise interference caused by uneven scanning light field from the signal acquisition source.

[0068] A laser confocal scanner is controlled to perform a row-by-row grid scan on a microarray protein chip, simultaneously acquiring the photoelectric conversion signal at each scan point. Dynamic time-series compensation is performed to address the photo-induced attenuation effect during the scanning process, resulting in the original pixel dataset. Specifically, after completing scanner parameter calibration and excitation power locking, the laser confocal scanner is controlled to continuously scan the entire surface of the microarray protein chip in a row-by-row grid scan mode. During the scan, the optical signal at each scan point on the chip surface is simultaneously acquired, and the optical signal is converted into a corresponding electrical signal by the photoelectric conversion unit, forming an initial photoelectric conversion signal sequence. To address the photo-induced attenuation effect caused by continuous laser irradiation and sensor response attenuation during the scan, a dynamic time-series compensation algorithm is used to correct the initial photoelectric conversion signal. The compensation calculation process is as follows: ,in, The photoelectric conversion signal after compensation at time t is given. The original photoelectric conversion signal was directly acquired at time t. This is the preset light-induced attenuation compensation coefficient. This represents the time difference between the current scanning time and the scanning start time. This dynamic compensation eliminates signal distortion caused by light-induced attenuation, ultimately yielding a complete and attenuation-free original pixel dataset.

[0069] The original pixel dataset is mapped to a two-dimensional pixel grid according to physical scan coordinates. After performing dark current subtraction and photoelectric sensor background noise calibration, it is converted into a signal matrix to be processed, represented by grayscale values. Specifically, after acquiring the original pixel dataset, each pixel data in the original pixel dataset is mapped one by one to a two-dimensional pixel grid that matches the physical size of the chip, according to the physical scan coordinate rules of the microarray protein chip. The horizontal physical coordinates of the two-dimensional pixel grid are as follows: The vertical physical coordinates are as follows This ensures that the pixel data accurately corresponds to the actual position of the chip; dark current subtraction and photoelectric sensor background noise calibration are performed sequentially. The calculation process for dark current subtraction is as follows: ,in, The pixel voltage value after deducting dark current. This is the original pixel voltage value. This represents the inherent dark current value of the photoelectric sensor. The sensor sampling resistor; after completing the dark current subtraction, the background noise calibration is performed. The calibration calculation process is as follows: ,in, The calibrated pixel voltage value. This represents the inherent noise level of the photoelectric sensor. After calibration, the calibrated pixel voltage values ​​are converted into grayscale values ​​in the 0-255 range according to standard grayscale conversion rules. All grayscale values ​​are arranged and encapsulated according to the coordinate order of a two-dimensional pixel grid, ultimately forming a signal matrix to be processed represented by grayscale values. Each element in the matrix Uniquely corresponds to coordinates in a two-dimensional pixel grid The grayscale values ​​at the location are used to construct the signal matrix to be processed.

[0070] In this embodiment of the invention, the technical means of calibrating the laser confocal scanner parameters and locking the preset excitation power according to the excitation spectral characteristics of the fluorescent probe, performing line-by-line grid scanning and dynamic time-series compensation for photo-induced attenuation effect, mapping the original pixel dataset to the grid according to physical coordinates and performing dark current subtraction and sensor background noise calibration overcome the technical problems of scanning parameter mismatch, signal distortion caused by photo-induced attenuation, and interference of dark current and background noise with the original signal acquisition, thereby obtaining accurate, stable, and low-noise interference original fluorescence images and constructing a standardized signal matrix to be processed.

[0071] In a preferred embodiment of the present invention, three pre-cured reference entities on the chip substrate are located from the signal matrix to be processed. The three reference entities are located in the non-array regions of the upper left corner, center point, and lower right corner of the chip, respectively. Each reference entity is composed of inert fluorescent microspheres of known concentration. The average fluorescence intensity of the local window where the three reference entities are located is extracted as discrete sampling values ​​of spatial heterogeneous background noise, including:

[0072] Based on the physical layout coordinate system of the microarray protein chip, the geometric centers of three preset non-array regions—the upper left corner, the center point, and the lower right corner—are retrieved and locked in the signal matrix to be processed. Three reference entities formed by the solidification of inert fluorescent microspheres of known concentration are identified. Specifically, this involves: mapping the two-dimensional pixel coordinates of the signal matrix to be processed to the chip's physical layout coordinates according to the preset physical layout coordinate system of the microarray protein chip; retrieving and locking the geometric center coordinates of the three non-array regions—the upper left corner, the center point, and the lower right corner—in the signal matrix according to preset position rules. These three non-array regions are all far from the protein array dot matrix region and are not subject to interference from the target protein fluorescence signal; and identifying the reference entities formed by the solidification of inert fluorescent microspheres of known concentration at the locked geometric center positions, thus completing the precise positioning of the three reference entities.

[0073] Using the geometric center of each reference entity as the origin, rectangular local windows of a preset size are extracted from the signal matrix to be processed. The boundaries of the local windows are strictly limited to the non-array area and maintain a preset safe distance from the edge of the microarray. Specifically, this involves: using the geometric center of each reference entity as the origin, and extracting a corresponding rectangular local window from the signal matrix to be processed according to preset size parameters; the boundaries of all extracted local windows are strictly limited to the non-array area of ​​the chip, while maintaining a preset safe distance between the edge of the local window and the edge of the microarray protein dot matrix to avoid the local window covering the effective signal area of ​​the array, thus eliminating the interference of the effective protein fluorescence signal on the background noise sampling results at the sampling area level and ensuring the purity of the sampling data.

[0074] The process involves iterating through all pixels covered within each local window, extracting their corresponding grayscale values, removing outliers that exceed a preset dynamic threshold range, and calculating the arithmetic mean of the remaining valid pixel values. Specifically, this includes: iterating through all pixels covered within each local window, extracting the grayscale value for each pixel, comparing the extracted grayscale value with a preset dynamic threshold range, removing outliers that exceed the threshold range, and retaining valid pixels that meet the threshold requirements; and then performing an arithmetic mean calculation on the remaining valid pixel values. The calculation process for the arithmetic mean is as follows: ;

[0075] in, It is the arithmetic mean of the grayscale values ​​of the effective pixels within the local window. This represents the total number of valid pixels within the local window after removing outliers. For the first in the local window The grayscale value corresponding to each effective pixel is calculated to obtain the average background noise intensity of each local window, thus eliminating the influence of single-point pixel anomalies on the sampling results.

[0076] The arithmetic mean of the gray values ​​corresponding to the three local windows is bound to the two-dimensional spatial coordinates of each reference entity to obtain discrete sampled values ​​characterizing the intensity of the spatial heterogeneous background noise distribution on the chip surface. Specifically, this involves binding the calculated arithmetic mean of the gray values ​​of the three local windows to the two-dimensional spatial coordinates of the corresponding reference entities, so that each noise intensity mean carries unique spatial location information. The size optimization stress constraint algorithm uses the geometric dimension parameters of the chip reference entity as the optimization design variable and the allowable stress of the chip material and the uniformity of structural stress distribution as the core constraints. It establishes a mechanical optimization algorithm that maps the entity size to the internal stress distribution through iterative calculation. Based on this algorithm, mechanical stress boundary constraints are applied to the bound coordinate-gray value data to eliminate abnormal noise intensity values ​​that exceed the stress bearing capacity of the entity size, ensuring that the correspondence between spatial coordinates and noise intensity conforms to the stress distribution law of the chip entity size. After the constraint correction is completed, three sets of discrete data containing spatial coordinates and fluorescence intensity mean are obtained. These discrete data are the discrete sampled values ​​characterizing the intensity of the spatial heterogeneous background noise distribution on the chip surface.

[0077] In this embodiment of the invention, by locating three inert fluorescent microsphere reference entities at the upper left corner, center point, and lower right corner of the non-array area of ​​the chip, extracting a local window that maintains a safe distance from the array dot matrix, removing outliers in grayscale, calculating the mean fluorescence intensity, and binding the mean value to the coordinates of the reference entities, the technical problems of background noise sampling being easily mixed with the effective array signal, insufficient sampling representativeness, distortion of sampling values ​​caused by single-point pixel anomalies, and inability to accurately characterize the spatial heterogeneous background noise distribution are overcome. Thus, pure, accurate, and spatially representative discrete background noise sampling values ​​are obtained.

[0078] In a preferred embodiment of the present invention, the two-dimensional coordinates of three reference entities and their corresponding average fluorescence intensity are used as control nodes. A virtual freeform surface with a continuous distribution of background noise across the entire chip surface is constructed using a bicubic interpolation algorithm, including:

[0079] To construct a control node dataset containing three-dimensional attributes—lateral physical coordinates, longitudinal physical coordinates, and mean fluorescence intensity—the discrete sampled values ​​and their associated two-dimensional spatial coordinates are spatially aligned and dimensionally normalized. Specifically, this involves: preprocessing the three sets of discrete sampled values; performing spatial alignment on the two-dimensional spatial coordinates bound to the discrete sampled values ​​to unify the origin and axial direction of the three sets of coordinates, eliminating data misalignment caused by coordinate system bias; performing dimensional normalization on the mean fluorescence intensity and physical coordinate parameters to convert physical quantities of different dimensions to a unified numerical range; and after completing spatial alignment and dimensional normalization, constructing a control node dataset containing three three-dimensional attributes: normalized lateral physical coordinates, normalized longitudinal physical coordinates, and normalized mean fluorescence intensity.

[0080] Based on the physical dimensions of the microarray protein chip and the pixel resolution of the original fluorescence image, a two-dimensional regular interpolation grid covering the entire chip surface is established. The control node dataset is then precisely mapped to the corresponding reference grid points in the two-dimensional regular interpolation grid. Specifically, this involves: calculating the number of horizontal and vertical grids in the two-dimensional regular interpolation grid based on the actual physical dimensions of the microarray protein chip and the pixel resolution of the original fluorescence image, ensuring that the grid size is consistent with the pixel size of the original fluorescence image, and establishing a two-dimensional regular interpolation grid that completely covers the entire surface of the microarray protein chip; and then precisely projecting the obtained control node dataset to the corresponding reference grid point positions in the two-dimensional regular interpolation grid according to the mapping relationship between normalized coordinates and interpolation grid coordinates, so that the three control nodes are bound one-to-one with the fixed points on the interpolation grid, ensuring that the starting node position of the interpolation calculation is accurate and providing stable constraint nodes for bicubic interpolation.

[0081] Using each reference grid point as the interpolation center, a cubic polynomial basis function is constructed in the neighborhood using a bicubic interpolation algorithm. By forcibly matching the continuity of function values ​​at the nodes and ensuring a smooth transition between the first and second partial derivatives, the fluorescence intensity fitting values ​​of all unfilled grid points in the two-dimensional regular interpolation grid are iteratively calculated. Specifically, this includes: using each reference grid point on the two-dimensional regular interpolation grid as the interpolation center, constructing a cubic polynomial basis function in the neighborhood of each reference grid point using a bicubic interpolation algorithm. The core polynomial expression for bicubic interpolation is: ;

[0082] in, Coordinates within the interpolation grid The fluorescence intensity fitting value at that location, Let be the coefficients of the polynomial to be solved. for coordinates Power of 1 for coordinates Power of 1 and As the index variable; during the interpolation calculation, the continuity condition of the function value at the reference grid point is forcibly matched, while ensuring a smooth transition between the first and second partial derivatives at the reference grid point. The polynomial coefficients are solved through an iterative optimization algorithm. Based on the solved polynomial coefficients, all unfilled grid points in the two-dimensional regular interpolation grid are traversed, and the fluorescence intensity fitting value at the corresponding position is calculated point by point to obtain fluorescence intensity fitting data covering the entire interpolation grid.

[0083] The fluorescence intensity fitting values ​​are reconstructed in three dimensions along the axes of the two-dimensional regular interpolation grid to obtain a virtual freeform surface for background noise that covers the entire chip surface and has continuous and smooth curvature. Specifically, this involves: reconstructing the calculated full-grid fluorescence intensity fitting values ​​in three dimensions along the horizontal and vertical axes of the two-dimensional regular interpolation grid, using the horizontal coordinate of the two-dimensional interpolation grid as the X-axis, the vertical coordinate as the Y-axis, and the fluorescence intensity fitting value as the Z-axis to construct a three-dimensional spatial surface model; through surface smoothing optimization, eliminating local curvature abrupt changes generated during the interpolation process, and finally fitting a virtual freeform surface for background noise that completely covers the entire surface of the microarray protein chip and has continuous and smooth curvature; this virtual freeform surface can realistically reproduce the nonlinear fluctuation morphology of spatial heterogeneous background noise from the center to the corners of the chip surface.

[0084] In this embodiment of the invention, by performing spatial alignment and dimensional normalization on discrete sampled values ​​and two-dimensional spatial coordinates, constructing a full-chip two-dimensional regular interpolation grid based on chip physical size and image resolution and accurately mapping control nodes, using bicubic interpolation algorithm to ensure continuous smoothness of function values ​​and partial derivatives of each order, and then reconstructing through three-dimensional space dimensionality upgrade, the technical problems of inconsistent data dimensions, mismatch between interpolation grid and chip physical scene, discontinuous and unsmooth background noise fitting surface, and difficulty in accurately representing the nonlinear spatial heterogeneous background noise distribution of the entire chip are overcome. Thus, a virtual free surface that covers the entire chip surface, has continuous and smooth curvature, and can truly reflect the background undulation law is constructed.

[0085] like Figure 2 As shown, in another preferred embodiment of the present invention, isotherms are divided under Riemannian metric on the virtual freeform surface, and multiple annular and radial regions are segmented along the principal curvature direction of the surface; the local gradient tensor of background noise is calculated in the annular and radial regions to obtain the normal offset correction coefficient of each pixel position relative to the virtual freeform surface, including:

[0086] A virtual freeform surface is introduced into the Riemann metric space. The first and second fundamental form parameters at each grid point on the surface are calculated to obtain the geometric curvature feature field characterizing the local undulations of the surface. Specifically, this involves: introducing the constructed background noise virtual freeform surface into the Riemann metric space; based on the two-dimensional parametric coordinates of the surface; traversing all interpolated grid points on the surface; and sequentially calculating the first and second fundamental form parameters at each grid point to quantify the local geometric morphology of the virtual freeform surface; the first fundamental form describes the intrinsic metric properties of the surface, and its expression is: ,in, , , Let be the coefficients of the first fundamental form, and be the inner product of the partial derivatives of the surface parameters. , The coordinates of the virtual freeform surface are two-dimensional parametric coordinates.

[0087] The second fundamental form is used to describe the extrinsic bending properties of a surface, and its expression is: ,in, , , The coefficients of the second fundamental form are obtained by the inner product of the surface normal vector and the second-order partial derivatives of the parameters;

[0088] Based on the coefficients of the first and second basic forms, the principal curvature, Gaussian curvature, and mean curvature parameters at each grid point are calculated. The curvature parameters of all grid points are arranged according to the topological structure of the interpolation grid to form a complete geometric curvature feature field that characterizes the local undulations of the virtual freeform surface.

[0089] Based on the geometric curvature feature field, the principal direction field of the virtual freeform surface is solved. Isothermal topological partitioning is performed along the trajectory of the principal curvature extreme values, dividing the virtual freeform surface into several annular regions with the same curvature gradient and radial regions radiating along the reference center. Specifically, based on the previously obtained geometric curvature feature field containing parameters such as principal curvature and Gaussian curvature of each grid node of the virtual freeform surface, data preprocessing is performed. All grid nodes are traversed to extract the two-dimensional physical coordinates, principal curvature values, and Gaussian curvature values ​​of each node. Abnormal curvature nodes caused by interpolation errors are removed, resulting in a coordinate-curvature pair for each grid node. The standardized curvature feature dataset is generated according to the relationship. A 2D quadtree is constructed in the two-dimensional physical coordinate space of the virtual freeform surface. The root node range is the entire two-dimensional physical coordinate domain of the virtual freeform surface. All standardized curvature feature data are included in the root node and the relevant parameters are stored. It is determined whether the number of curvature nodes in the root node exceeds the preset threshold. If it does, it is divided into four child nodes along the midpoint of the x-axis and y-axis and the corresponding curvature nodes are assigned. This division operation is repeated until the number of nodes in all child nodes does not exceed the preset threshold. The coordinate boundary of each terminal child node is marked, the internal curvature node parameters are stored, and a mapping relationship with the physical coordinates is established.

[0090] The quadtree range query algorithm is used to locate the trajectory of principal curvature extreme values. Based on the distribution law of curvature feature field, the threshold of principal curvature extreme values ​​is set. The query is recursively performed starting from the root node, skipping nodes with no intersection. The terminal child nodes that meet the conditions are selected, and their two-dimensional physical coordinates are recorded. The coordinates are then linearly interpolated and fitted to form a continuous trajectory of principal curvature extreme values. Based on this trajectory, the tangent direction of each point is analyzed. Through vector orthogonalization, two sets of mutually orthogonal direction vectors are obtained, namely the tangent direction of the trajectory with the most significant principal curvature change and the normal direction of the smoothest trajectory. This determines the principal direction field of the virtual freeform surface.

[0091] Using the reference entity at the center of the surface as the radiation origin, isotherms are drawn by extracting fluorescence intensity distribution data from the virtual freeform surface. Regions with consistent curvature gradients are grouped according to the curvature gradient information stored in the quadtree. Multiple annular regions with uniform curvature gradients are divided along the principal curvature extreme value trajectory. At the same time, multiple radial regions are divided along the principal direction field radiation direction. All region boundaries are verified by quadtree range query to ensure that they do not exceed the virtual freeform surface range and do not cross curvature gradient abrupt changes. Finally, annular and radial regions are formed that intersect each other, completely cover the entire virtual freeform surface, and have consistent background noise fluctuations in each region.

[0092] Within each annular and radial region, using the fluorescence intensity distribution of the virtual freeform surface as a scalar field, the partial derivatives of the intensity values ​​in the horizontal and vertical coordinates are calculated pixel-by-pixel to construct a local gradient tensor matrix characterizing the spatial rate of change and directional characteristics of background noise. Specifically, this includes: using the fluorescence intensity fitted value of the virtual freeform surface as a scalar field, calculating the spatial partial derivatives of the fluorescence intensity values ​​pixel-by-pixel within each of the divided annular and radial regions to construct a local gradient tensor matrix characterizing the spatial rate of change and directional characteristics of background noise; and traversing the target pixels within the region to calculate the fluorescence intensity in the horizontal coordinate... with vertical coordinate The first-order partial derivative is calculated using the following formula: , .in, pixel coordinates The fluorescence intensity fitting value at that location, , Given a unit step size for the pixel grid, the horizontal and vertical partial derivatives of the same pixel are combined into a two-dimensional gradient vector. Then, the gradient vectors of all pixels in the region are matrixed and encapsulated in the order of physical coordinates to obtain the local gradient tensor matrix. This matrix can completely quantify the spatial change rate and directional characteristics of background noise in each partition.

[0093] The local gradient tensor matrix and the unit normal vector of the virtual freeform surface at each pixel coordinate are subjected to vector projection operation. The intensity deviation of each pixel along the surface normal direction is extracted, and after linear scaling, the normal offset correction coefficient of each pixel position relative to the virtual freeform surface is obtained. Specifically, this includes: solving for the unit normal vector of the virtual freeform surface at each pixel coordinate. The unit normal vector is obtained by cross product of the partial derivatives of the surface parameters and normalization; the gradient vector of the corresponding pixel in the constructed local gradient tensor matrix is ​​then used. , with unit normal vector Perform vector dot product projection to extract the intensity deviation of each pixel along the surface normal direction. The projection formula is as follows: ,in, This represents the intensity deviation of the current pixel along the surface normal direction.

[0094] Linear scaling is applied to the intensity deviation of all pixels, mapping the deviation to a numerical range that matches the grayscale values ​​of the original fluorescence image. The linear scaling formula is as follows: ,in, pixel coordinates Normal offset correction factor at the location, , These are the minimum and maximum values ​​of the global intensity deviation. This is the preset range of correction coefficient values.

[0095] Through the above calculations, the normal offset correction coefficient of each pixel position relative to the virtual freeform surface is finally obtained. This coefficient can accurately characterize the degree of background noise deviation of each pixel.

[0096] In this embodiment of the invention, by introducing a virtual freeform surface into the Riemann metric space to calculate the geometric curvature feature field, performing isotherm topological partitioning based on the principal curvature extreme value trajectory to obtain annular and radial regions, constructing a local gradient tensor matrix representing the spatial variation characteristics of noise pixel by pixel, and obtaining the normal offset correction coefficient through vector projection and linear scaling, the technical problems of being unable to accurately quantify the local undulation shape of the surface, unreasonable noise region partitioning, difficulty in representing the spatial variation characteristics of background noise, and inability to obtain pixel-level accurate correction parameters are overcome. Thus, the fine adaptive partitioning of the background noise region is achieved, the noise deviation degree of each pixel position is accurately quantified, and a stable and reliable pixel-level normal offset correction coefficient is obtained.

[0097] In a preferred embodiment of the present invention, within each annular and radial region, using the fluorescence intensity distribution of the virtual freeform surface as a scalar field, the partial derivatives of the intensity values ​​in the horizontal and vertical coordinates are calculated pixel-by-pixel to construct a local gradient tensor matrix characterizing the spatial rate of change and directional features of background noise, including:

[0098] The target pixels within each annular and radial region are traversed. Based on the grid topology of the virtual freeform surface, the horizontal and vertical position indices of each target pixel in the two-dimensional pixel coordinate system, as well as the corresponding fluorescence intensity fitting value, are read. Specifically, this includes: based on the grid topology of the virtual freeform surface, a full traversal is performed on all the divided annular and radial regions, and each target pixel in each region is extracted sequentially; during the traversal, the horizontal and vertical position indices of each target pixel in the two-dimensional pixel coordinate system, as well as the corresponding fluorescence intensity fitting value on the virtual freeform surface, are read simultaneously to ensure that the read position indices and fluorescence intensity fitting values ​​correspond one-to-one with the grid nodes of the virtual freeform surface and are completely matched with the physical coordinate system of the microarray protein chip.

[0099] Using a preset discrete difference operator template, the center difference value of each target pixel in the horizontal and vertical coordinate axes is calculated respectively. The horizontal and vertical partial derivative components that characterize the rate of change of background noise intensity along the horizontal and vertical axes are extracted. Specifically, the process includes: using a preset discrete difference operator template, performing a center difference operation on each target pixel, solving for the rate of change of background noise intensity in the horizontal and vertical coordinate axes respectively, and obtaining the horizontal and vertical partial derivative components.

[0100] The calculation process for the horizontal center difference and horizontal partial derivative components of the target pixel is as follows: The calculation process for the longitudinal center difference and longitudinal partial derivative components of the target pixel is as follows: ,in, This represents the lateral partial derivative component of the target pixel. This represents the longitudinal partial derivative component of the target pixel. The fluorescence intensity fitting value of the target pixel. The fluorescence intensity fitted value of the pixel adjacent to the right horizontally of the target pixel. The fluorescence intensity fitted value of the pixel horizontally to the left of the target pixel. The fluorescence intensity fitted value is the value of the pixel adjacent to the vertically above the target pixel. The fluorescence intensity fitting value is the value of the adjacent pixel vertically below the target pixel. Through the above calculation, the spatial change rate of background noise intensity in the horizontal and vertical directions can be accurately extracted, local pixel noise interference can be eliminated, and the accuracy of the partial derivative component can be guaranteed.

[0101] The horizontal and vertical partial derivative components corresponding to the same target pixel are orthogonally combined to construct a two-dimensional gradient vector containing information about the magnitude of the change and the spatial gradient direction. Specifically, this involves orthogonally combining the horizontal and vertical partial derivative components corresponding to the same target pixel to construct a two-dimensional gradient vector that simultaneously represents the magnitude of the noise intensity change and the spatial gradient direction. The formula for constructing the two-dimensional gradient vector is as follows: ,in, The target pixel is a two-dimensional gradient vector. This vector integrates the horizontal and vertical partial derivative features in an orthogonal form, contains the amplitude information of the change in background noise intensity at this pixel, and also clarifies the spatial change direction of noise intensity, thus fully preserving the spatial distribution characteristics of local background noise.

[0102] The two-dimensional gradient vectors corresponding to all target pixels in each annular and radial region are matrix-encapsulated according to their original physical coordinate order to obtain a local gradient tensor matrix covering the entire target region. Specifically, this involves: arranging the two-dimensional gradient vectors corresponding to all target pixels in each annular and radial region in an ordered manner according to the original physical coordinate order of the microarray protein chip, converting the single-dimensional gradient vectors into a two-dimensional matrix for encapsulation; the horizontal dimension of the matrix corresponds one-to-one with the horizontal position index of the two-dimensional pixel coordinate system, and the vertical dimension of the matrix corresponds one-to-one with the vertical position index of the two-dimensional pixel coordinate system. Each element in the matrix is ​​a two-dimensional gradient vector at the corresponding coordinate position, ultimately generating a local gradient tensor matrix covering all annular and radial regions; this matrix can completely characterize the spatial rate of change and directional characteristics of background noise in each region.

[0103] In this embodiment of the invention, by traversing the target pixels in each region to read the position index and fluorescence intensity fitting value, calculating the horizontal and vertical center differences using a discrete difference operator template to obtain the partial derivative components, orthogonally combining the partial derivative components to construct a two-dimensional gradient vector, and then encapsulating it according to the original physical coordinate matrix, the technical problem of difficulty in accurately extracting the spatial change rate of background noise, inability to quantify the amplitude and direction features of noise change, and distortion of noise representation caused by incomplete gradient information is overcome. Thus, a local gradient tensor matrix that comprehensively and accurately represents the spatial change rate and direction features of background noise is constructed.

[0104] In a preferred embodiment of the present invention, a preliminary background suppression image is obtained by performing a point-by-point difference operation between the normal offset correction coefficient and the gray value of each corresponding pixel in the original fluorescence image, including:

[0105] The normal offset correction coefficients are received. Based on the physical coordinate system mapping relationship between the signal matrix to be processed and the virtual freeform surface mesh, the coordinates of each node in the normal offset correction coefficient matrix are spatially registered and aligned with the pixel coordinates of the signal matrix to be processed, establishing a one-to-one pixel-level mapping index. Specifically, this includes: receiving the previously constructed normal offset correction coefficients, which are derived from the gradient tensor matrices of the annular and radial regions and are the core foundation for subsequent difference operations; and performing coordinate registration operations based on the physical coordinate system mapping relationship between the pixel matrix corresponding to the original fluorescence image and the virtual freeform surface mesh to ensure that the coordinate systems of the two are unified and without deviation. The core of the specific operation is to establish the correspondence between the normal offset correction coefficient matrix and the pixel coordinates of the signal matrix to be processed: Let the pixel coordinates of the signal matrix to be processed be... ,in For horizontal pixel indexing, The vertical pixel index is used; the node coordinates of the normal offset correction coefficient matrix are... ,in To correct the horizontal index of the coefficient matrix, This is a vertical index. Through coordinate translation operations, it becomes... , This achieves spatial alignment of the two matrices and establishes a one-to-one pixel-level mapping index, ensuring that each original fluorescent pixel can be matched with a unique normal offset correction coefficient during subsequent point-by-point difference operations, thus avoiding deviations in the difference results caused by coordinate misalignment.

[0106] Based on pixel-level mapping indexes, the system traverses the aligned pixel index positions and simultaneously reads the initial grayscale value of the corresponding pixel in the original fluorescence image and the normal offset correction coefficient matching the position, completing the data alignment before the difference operation. Specifically, based on the established pixel-level mapping indexes, it begins traversing all pixel positions of the signal matrix to be processed and the correction coefficient matrix to achieve full data alignment before the difference operation. The traversal process strictly follows the original physical coordinate order of the microarray protein chip, from left to right horizontally and from top to bottom vertically, completing two core operations pixel by pixel: first, reading the initial grayscale value of the corresponding pixel in the original fluorescence image, denoted as... ,in For horizontal pixel indexing, Vertical pixel index, The first is the original fluorescence intensity of the pixel containing background noise; the second is the normal offset correction coefficient corresponding to the pixel position, denoted as... This coefficient is calculated from the local gradient tensor matrix in the previous stage and corresponds to the intensity value of the background noise. The read data is verified in real time. If a pixel does not have a corresponding correction coefficient, the average correction coefficient of the four adjacent pixels is used to supplement it. The formula is as follows: This ensures that each pixel has a corresponding correction coefficient, preventing data loss from interrupting the differential operation.

[0107] The normal offset correction coefficient and the corresponding initial grayscale value are subjected to a point-by-point algebraic difference operation to remove intensity superposition redundancy caused by spatially heterogeneous background noise, resulting in a denoised intermediate grayscale dataset carrying effective signal features. Specifically, for each pixel, a point-by-point algebraic difference operation is performed between the normal offset correction coefficient and the original fluorescence grayscale value. This operation removes spatially heterogeneous background noise from the original fluorescence signal, including noise generated by substrate autofluorescence, scanning light field inhomogeneity, and edge scattering, while retaining effective low-abundance protein signals. The specific calculation formula is as follows:

[0108] ,in, This is the grayscale value of the pixel after the difference operation, i.e., the grayscale value after denoising. The initial grayscale value of this pixel in the original fluorescence image includes background noise. This is the normal offset correction coefficient corresponding to the pixel, which corresponds to the intensity of the background noise. During the calculation, the principle of pixel-by-pixel traversal is strictly followed. The calculation of each pixel is performed independently without cross-interference. The specific operation is as follows: First, read a certain pixel... of and Substituting into the formula, we can calculate the result. After recording the result, the operation continues for the next pixel until the difference calculation for all pixels is completed. The core advantage of this operation is that, unlike traditional fixed-threshold subtraction, it can adjust the subtraction based on the background noise intensity of each pixel. This process allows for precise subtraction, avoiding the loss of effective signals due to global subtraction, while also eliminating the redundancy of background noise intensity superposition, thus highlighting low-abundance protein signals.

[0109] For negative gray-level outliers below the zero threshold generated by the difference operation in the denoised intermediate gray-level dataset, truncation and zeroing are performed. The processed effective gray-level values ​​are then rearranged and encapsulated according to the original two-dimensional pixel grid to obtain a preliminary background suppression image. Specifically, after the difference operation, some negative gray-level outliers below the zero threshold are generated. These outliers are caused by excessive background noise removal; retaining them would affect the accuracy of subsequent detection. Therefore, it is necessary to truncate and zero-set the denoised intermediate gray-level dataset, i.e., all... The dataset is composed of data points, and outlier handling and repackaging are performed. First, the grayscale value zero threshold is set to 0 for all... Perform filtering: When When <0, perform truncation to zero, setting the grayscale value of the pixel to 0; when When the grayscale value is ≥0, the grayscale value is retained to avoid interference from negative grayscale values ​​on the effective signal. All processed values ​​are then processed according to the original physical coordinate order of the microarray protein chip. The image is rearranged and encapsulated to restore a two-dimensional pixel matrix with the same size as the original fluorescence image. This matrix is ​​the initial background suppression image, which has eliminated most of the spatially heterogeneous background noise and preserved the spatial continuity of the effective signal.

[0110] In this embodiment of the invention, the technical means of performing pixel-level spatial registration and alignment of the normal offset correction coefficient matrix and the signal matrix to be processed based on the physical coordinate system mapping relationship, establishing a one-to-one correspondence mapping index, performing algebraic difference operations point by point to remove spatial heterogeneous background noise intensity redundancy, and truncating and setting negative gray-level outliers to zero before re-encapsulating them according to the original grid, thus overcoming the technical problems of misalignment between correction coefficients and original pixel coordinates, mismatch of difference operation data, inability to accurately remove spatial heterogeneous background noise, and abnormal negative values ​​generated by difference interfering with effective signals, thereby achieving pixel-level accurate background suppression and obtaining a preliminary background suppression image that retains effective signal features and is free from abnormal data interference.

[0111] In a preferred embodiment of the present invention, anisotropic diffusion filtering is performed on the residual light scattering noise at the edges of each array point in the preliminary background suppression image, and the final output is a clean fluorescence signal image with suppressed background noise, including:

[0112] Based on a virtual freeform surface, Gaussian curvature values ​​of each grid node are extracted. These Gaussian curvature values ​​are then mapped to the pixel array of the signal matrix to be processed using physical coordinates, resulting in a Gaussian curvature distribution parameter map characterizing the spatial fluctuation of background noise. Specifically, this includes: extracting Gaussian curvature values ​​and performing coordinate mapping operations based on the constructed virtual freeform surface for background noise, providing noise distribution characteristics for subsequent adaptive diffusion filtering; traversing all grid nodes of the virtual freeform surface; and calculating Gaussian curvature values ​​node-by-node based on the calculated first and second fundamental form coefficients. The core formula for calculating Gaussian curvature is: ,in, Let be the Gaussian curvature value of a mesh node on a virtual freeform surface. , , The first fundamental formal coefficient of this node characterizes the intrinsic metric properties of the surface. , , The second fundamental form coefficient of this node characterizes the intrinsic bending properties of the surface; the sign and magnitude of the Gaussian curvature value can accurately reflect the steepness of the local undulations of the virtual freeform surface, and thus indirectly characterize the spatial undulation law of the background noise in the corresponding region. The larger the absolute value of curvature, the more intense the background noise undulations in the region, and the more likely there is residual light scattering noise.

[0113] After calculating the Gaussian curvature of all grid nodes, based on the established physical coordinate system mapping relationship between the signal matrix to be processed and the virtual freeform surface grid, the Gaussian curvature value of each grid node is mapped one by one to the corresponding pixel array position of the signal matrix to be processed, according to its corresponding physical coordinates. This ensures that the Gaussian curvature value and the pixel coordinates correspond precisely without misalignment. All mapped Gaussian curvature values ​​are arranged according to the topology of the original two-dimensional pixel grid to form a complete Gaussian curvature distribution parameter map characterizing the spatial fluctuation law of background noise on the chip surface. This parameter map transforms the spatial distribution characteristics of background noise into quantifiable curvature parameters.

[0114] The process involves reading the initial background suppression image, identifying the physical boundaries of each array point using an edge contour extraction operator, cropping a transition zone pixel region of a preset width around the boundary, and constructing a local noise mask characterizing the spatial distribution of residual light scattering noise. Specifically, this includes reading the generated initial background suppression image, which has removed most of the spatially heterogeneous background noise, but still contains residual light scattering noise at the edges of the array points. This type of noise is concentrated in the transition region around the physical boundaries of the array points and requires precise filtering through target recognition and mask marking.

[0115] A preset edge contour extraction operator is used to perform a full-domain scan of the preliminary background suppression image to identify the physical boundaries of each array point. During the scanning process, by detecting the abrupt change characteristics of pixel grayscale values, the effective signal area and edge transition area of ​​the array point are distinguished, and the boundary contour of each array point is accurately locked to ensure that the boundary identification is complete and without false positives. Based on the identified physical boundaries of the array points, a transition zone pixel area of ​​preset width is extracted from the outer edge of the boundary. The width of the transition zone is preset according to the array point size and scanning resolution of the microarray protein chip to ensure that the transition zone completely covers all light scattering residual noise distribution areas, while not including the effective signal area inside the array point, so as to avoid destroying the effective signal during subsequent filtering.

[0116] All the captured transition zone pixel regions are marked to construct a local noise mask. In the mask, the marked pixel region transition zone corresponds to the light scattering residual noise distribution region, and the unmarked pixel region corresponds to the effective signal region without residual noise. Through this mask, the residual noise region and the effective signal region can be accurately distinguished.

[0117] The pixel region covered by the local noise mask is used as the diffusion iteration field. An anisotropic diffusion filtering algorithm is called to perform multiple rounds of pixel gray-level redistribution. In each iteration, the data at the corresponding position of the Gaussian curvature distribution parameter map is dynamically mapped to the adaptive regularization parameter of the diffusion control equation. The diffusion intensity and edge protection threshold are adjusted in real time according to the curvature gradient difference. Specifically, the pixel region covered by the constructed local noise mask is used as the diffusion iteration field. Anisotropic diffusion filtering operation is performed only on the pixels in this region to achieve targeted suppression of residual light scattering noise, while preserving the effective signal edge to the maximum extent.

[0118] An anisotropic diffusion filtering algorithm is invoked, and initial iteration parameters, including initial diffusion intensity, iteration step size, and maximum number of iterations, are set. Multiple rounds of pixel grayscale redistribution iterations are then initiated. The core of the iteration process is to dynamically adjust the adaptive regularization parameters of the diffusion control equation based on the Gaussian curvature distribution parameter map, achieving real-time adaptation between diffusion intensity and edge protection threshold. The core control equation of the anisotropic diffusion filter is: ,in, for After round of iterations, the coordinates ( The pixel grayscale value at () For iteration rounds; For divergence operators; for After round of iterations, the coordinates ( The pixel grayscale gradient at () location; for During round iteration, coordinates ( The adaptive regularization parameter at position () is obtained by dynamically mapping the Gaussian curvature value at the corresponding position on the Gaussian curvature distribution parameter map. The mapping formula is: ,in, coordinates The Gaussian curvature value at that location is derived from the generated Gaussian curvature distribution parameter map. This is a preset curvature adjustment coefficient used to control the variation range of the regularization parameter.

[0119] During each iteration, the diffusion intensity and edge protection threshold are adjusted in real time based on the difference in Gaussian curvature gradient: for regions with large absolute values ​​of Gaussian curvature, areas with drastic fluctuations in background noise, and areas with concentrated residual noise, adaptive regularization parameters are used. Smaller values ​​increase diffusion intensity, accelerating the gray-level redistribution of residual noise and achieving rapid noise suppression; for regions with small absolute values ​​of Gaussian curvature, the background noise is smooth and close to the edge region of the effective signal, thus enabling adaptive regularization parameters. A larger value reduces the diffusion intensity and increases the edge protection threshold to prevent the edges of the effective signal from being blurred. After each iteration, the grayscale value of the diffused pixels is verified to ensure that the grayscale value is within a reasonable range of 0-255 before proceeding to the next iteration, until the preset convergence condition is met.

[0120] When the diffusion iteration reaches the preset convergence condition or the pixel grayscale residual of adjacent iterations is lower than the preset threshold, the filtering operation is terminated. The effective grayscale values ​​after iterative convergence are repackaged according to the original two-dimensional pixel grid topology to obtain a pure fluorescence signal image with suppressed background noise. Specifically, this includes: during multiple diffusion iterations, the iteration convergence status is monitored in real time to determine whether to terminate the filtering operation. There are two termination conditions, and termination is possible if either one is met: 1) the diffusion iteration reaches the preset maximum number of iterations; 2) the pixel grayscale residual of two adjacent iterations is lower than the preset threshold. The formula for calculating the grayscale residual is: ,in, The average grayscale residual between two adjacent iterations. , These represent the number of pixels horizontally and vertically in the signal matrix to be processed, respectively. For the first Coordinates after round iteration The pixel grayscale value at that location, For the first Coordinates after round iteration Pixel grayscale value at the location; preset threshold Based on the chip detection accuracy requirements, when < When the iteration has converged and the filtering effect has stabilized, there is no need to continue iterating.

[0121] After the filtering operation terminates, all effective pixel gray values ​​after iterative convergence are extracted. These gray values ​​are rearranged and encapsulated according to the original two-dimensional pixel grid topology of the microarray protein chip, and restored to a two-dimensional pixel matrix with the same size and accurate coordinate correspondence as the original fluorescence image. This matrix is ​​the pure fluorescence signal image with background noise completely suppressed.

[0122] In this embodiment of the invention, the technical means of extracting Gaussian curvature from a virtual freeform surface and mapping it to a background noise fluctuation parameter map, identifying the array point boundaries to construct a local noise mask for the edge transition zone, dynamically adjusting the diffusion intensity and edge protection threshold with Gaussian curvature for adaptive anisotropic diffusion filtering, terminating the iteration according to the convergence condition and re-encapsulating the signal, overcome the technical problems of not being able to target and suppress residual light scattering noise at the array point edges, the fixed diffusion filtering parameters leading to unsatisfactory denoising or blurred effective signals, and the difficulty in balancing residual noise removal and edge signal protection. This achieves accurate adaptive suppression of edge scattering noise, fully preserves the effective signal of the array points, and finally outputs a pure fluorescence signal image with completely suppressed background noise and clear and accurate signal.

[0123] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the system as described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0124] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the system as described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0125] Experimental example:

[0126] I. Experimental Background

[0127] This experimental example validates background noise suppression for the detection of allergen-specific biomarkers using a protein chip. The chip employs laser confocal scanning to acquire fluorescence images, targeting low-abundance protein biomarkers. Due to spatially heterogeneous background noise interference from substrate autofluorescence, uneven scanning light field, and edge light scattering on the chip surface, traditional global mean subtraction methods are insufficient for accurate clinical detection. The aim of this experiment is to verify the effectiveness of the background noise suppression system proposed in this invention in improving the detection rate of low-abundance protein signals.

[0128] II. Experimental Equipment and Parameters

[0129] The experiment used the background noise suppression system described in this invention, and the main equipment included:

[0130] Laser confocal scanner: excitation wavelength 488nm, detection pinhole diameter 1.0 Airy Unit, laser excitation power locked at preset excitation power 2.5mW.

[0131] Microarray protein chip: Three reference entities are pre-immobilized on the chip substrate, located in the non-array areas of the upper left corner, center point and lower right corner of the chip, respectively. Each reference entity is composed of inert fluorescent microspheres of known concentration.

[0132] Image processing workstation: Equipped with a high-performance GPU for performing bicubic interpolation, Riemann metric calculations, and anisotropic diffusion filtering algorithms.

[0133] III. Experimental Procedures and Result Analysis

[0134] Step 1: Acquisition of raw fluorescence images

[0135] The raw fluorescence images of the microarray protein chip were acquired using a laser confocal scanner at a preset excitation power, and the raw fluorescence images were used as the signal matrix to be processed.

[0136] Based on the excitation spectral characteristics of the fluorescent probes labeled on the microarray protein chip, the excitation wavelength of the laser confocal scanner was calibrated to match the aperture of the detection pin, and the laser excitation power was locked at 2.5mW. The laser confocal scanner was controlled to perform row-by-row grid scanning of the microarray protein chip, synchronously acquiring the photoelectric conversion signal of each scanning point, and dynamically compensating for the photo-induced attenuation effect during the scanning process.

[0137] Figure 3 A simulation diagram of the chip reference entity arrangement is shown. The three reference entities are located in the non-array areas at the upper left corner, center point, and lower right corner of the chip, respectively, maintaining a preset safe distance from the edge of the microarray. This arrangement comprehensively characterizes the spatial heterogeneous background noise distribution on the chip surface from the center to the corners.

[0138] Step 2: Localization of the reference entity and sampling of background noise

[0139] Three pre-cured reference entities on the chip substrate are located from the signal matrix to be processed. The average fluorescence intensity of the local window where the three reference entities are located is extracted as the discrete sampling value of the spatial heterogeneous background noise.

[0140] Figure 4 The background noise sampling statistics for three reference entities are presented over 10 scan rounds. The noise intensity of the top-left reference entity fluctuates between 40 and 50 au, the center reference entity between 48 and 56 au, and the bottom-right reference entity between 44 and 54 au. The differences in noise intensity at these three locations reflect the spatial heterogeneity of the chip surface.

[0141] Step 4: Isotherm Division and Gradient Tensor Calculation

[0142] The virtual freeform surface is divided into isotherms under Riemannian metric, and multiple annular and radial regions are segmented along the principal curvature direction of the surface. The local gradient tensor of background noise is calculated in the annular and radial regions to obtain the normal offset correction coefficient of each pixel position relative to the virtual freeform surface.

[0143] Figure 5 The statistical results of the background noise gradient tensor for each region are presented. In the annular region, the gradient magnitude is between 4.0 and 4.5; in the radial region, the gradient magnitude is between 5.4 and 6.1, indicating that the spatial variation of background noise is more drastic in the radial region.

[0144] IV. Experimental Conclusions

[0145] The above experiments verify that the background noise suppression system for microarray protein chips proposed in this invention effectively suppresses spatially heterogeneous background noise and improves the detection rate of low-abundance protein signals. Experimental results show that:

[0146] By sampling from three reference points in the non-array region of the chip, pure, accurate, and spatially representative discrete background noise samples are obtained, avoiding interference from the effective array signal on the sampling results.

[0147] The virtual freeform surface constructed by the bicubic interpolation algorithm realistically reproduces the nonlinear undulation of spatial heterogeneous background noise on the chip surface from the center to the corners, with continuous and smooth curvature.

[0148] By using isotherm partitioning under Riemannian metric and local gradient tensor calculation, a refined adaptive partitioning of the background noise region is achieved, accurately quantifying the noise deviation at each pixel position.

[0149] Anisotropic diffusion filtering is used to specifically suppress residual light scattering noise at the edges of array points, while fully preserving the effective signals of the array points.

[0150] Compared with existing technologies, the method of the present invention improves the signal-to-noise ratio by 28dB, reduces the edge quantification bias to 5%, and improves the detection rate and detection repeatability of low-abundance protein signals.

[0151] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A background noise suppression system for microarray protein chips, characterized in that, include: The acquisition module is used to acquire the raw fluorescence image of the microarray protein chip using a laser confocal scanner at a preset excitation power, and to use the raw fluorescence image as a signal matrix to be processed. The extraction module is used to locate three pre-cured reference entities on the chip substrate from the signal matrix to be processed. The three reference entities are located in the non-array regions of the upper left corner, center point and lower right corner of the chip, respectively. Each reference entity is composed of inert fluorescent microspheres of known concentration. The average fluorescence intensity of the local window where the three reference entities are located is extracted as discrete sampling values ​​of spatial heterogeneous background noise. The module is used to construct a virtual freeform surface with a continuous distribution of background noise on the entire chip surface using a bicubic interpolation algorithm, with the two-dimensional coordinates of three reference entities and the corresponding mean fluorescence intensity as control nodes. The partitioning module is used to perform isotherm partitioning of the virtual freeform surface under Riemannian metric, dividing the surface into multiple annular and radial regions along the principal curvature direction; The local gradient tensor of the background noise is calculated in the annular and radial regions to obtain the normal offset correction coefficient of each pixel position relative to the virtual freeform surface; The calculation module is used to perform point-by-point difference operations between the normal offset correction coefficient and the gray value of each corresponding pixel in the original fluorescence image to obtain a preliminary background suppression image; The processing module is used to perform anisotropic diffusion filtering on the residual light scattering noise at the edges of each array point in the preliminary background suppression image, and finally outputs a pure fluorescence signal image with suppressed background noise.

2. The background noise suppression system for microarray protein chips according to claim 1, characterized in that, Raw fluorescence images of the microarray protein chip were acquired using a laser confocal scanner at a preset excitation power, and these raw fluorescence images were used as a signal matrix to be processed, including: Based on the excitation spectral characteristics of the fluorescent probes labeled on the microarray protein chip, the excitation wavelength of the laser confocal scanner and the aperture of the detection pin are calibrated to match the parameters, and the laser excitation power is locked to the preset excitation power; The laser confocal scanner is controlled to perform row-by-row grid scanning on the microarray protein chip, and the photoelectric conversion signal of each scanning point is collected synchronously. Dynamic time-series compensation is performed on the photo-induced attenuation effect during the scanning process to obtain the original pixel dataset. The original pixel dataset is mapped to a two-dimensional pixel grid according to the physical scan coordinates. After performing dark current subtraction and photoelectric sensor background noise calibration, it is converted into a signal matrix to be processed, represented by grayscale values.

3. The background noise suppression system for microarray protein chips according to claim 2, characterized in that, Three pre-cured reference entities on the chip substrate are located from the signal matrix to be processed. These three reference entities are located in the non-array regions at the upper left, center, and lower right corners of the chip, respectively. Each reference entity is composed of inert fluorescent microspheres of known concentration. The average fluorescence intensity of the local window containing each of the three reference entities is extracted as discrete sampling values ​​of spatially heterogeneous background noise, including: Based on the physical layout coordinate system of the microarray protein chip, the geometric centers of three preset non-array regions—the upper left corner, the center point, and the lower right corner—are retrieved and locked in the signal matrix to be processed, and three reference entities formed by the solidification of inert fluorescent microspheres of known concentration are identified. Using the geometric center of each reference entity as the origin, a rectangular local window of a preset size is extracted from the signal matrix to be processed. The boundary of the local window is strictly limited to the non-array area and maintains a preset safe distance from the edge of the microarray. Iterate through all pixels covered in each local window, extract the corresponding grayscale values ​​and remove outliers that exceed the preset dynamic threshold range, and calculate the arithmetic mean of the grayscale values ​​of the remaining valid pixels. The arithmetic mean of the gray values ​​corresponding to the three local windows is bound to the two-dimensional spatial coordinates of each reference entity to obtain discrete sampled values ​​that characterize the intensity of the spatial heterogeneous background noise distribution on the chip surface.

4. The background noise suppression system for microarray protein chips according to claim 3, characterized in that, Using the two-dimensional coordinates of three reference entities and their corresponding mean fluorescence intensities as control nodes, a virtual freeform surface with a continuous background noise distribution across the entire chip surface is constructed using a bicubic interpolation algorithm, including: The discrete sampled values ​​and their bound two-dimensional spatial coordinates are spatially aligned and normalized to construct a control node dataset containing three-dimensional attributes: horizontal physical coordinates, vertical physical coordinates, and mean fluorescence intensity. Based on the physical size of the microarray protein chip and the pixel resolution of the original fluorescence image, a two-dimensional regular interpolation grid covering the entire chip surface is established, and the control node dataset is accurately mapped to the corresponding reference grid points in the two-dimensional regular interpolation grid. Using each reference grid point as the interpolation center, a cubic polynomial basis function is constructed in the neighborhood through a bicubic interpolation algorithm. By forcing the continuity of function values ​​at matching nodes and the smooth transition condition between first-order and second-order partial derivatives, the fluorescence intensity fitting values ​​of all unfilled grid points in the two-dimensional regular interpolation grid are iteratively calculated. The fluorescence intensity fitting values ​​are reconstructed in three dimensions along the axis of the two-dimensional regular interpolation grid to obtain a virtual freeform surface with background noise that covers the entire chip surface and has continuous and smooth curvature.

5. The background noise suppression system for microarray protein chips according to claim 4, characterized in that, The virtual freeform surface is divided into isotherms under the Riemannian metric, and multiple annular and radial regions are segmented along the principal curvature direction of the surface. The local gradient tensor of the background noise is calculated in the annular and radial regions to obtain the normal offset correction coefficient of each pixel position relative to the virtual freeform surface, including: By introducing a virtual freeform surface into the Riemann metric space, the parameters of the first and second fundamental forms at each grid point on the surface are calculated to obtain the geometric curvature feature field that characterizes the local undulation of the surface. The principal direction field of the virtual freeform surface is solved based on the geometric curvature characteristic field. Isotherm topological partitioning is performed along the trajectory of the extreme value change of the principal curvature, dividing the virtual freeform surface into several annular regions with the same curvature gradient and radial regions radiating along the reference center. Within each annular and radial region, the fluorescence intensity distribution of the virtual freeform surface is used as a scalar field. The combination of partial derivatives of the intensity value in the horizontal and vertical coordinates is calculated pixel by pixel to construct a local gradient tensor matrix that characterizes the spatial rate of change and directional features of background noise. The local gradient tensor matrix and the unit normal vector of the virtual freeform surface at each pixel coordinate are subjected to vector projection operation. The intensity deviation of each pixel along the normal direction of the surface is extracted. After linear scaling, the normal offset correction coefficient of each pixel position relative to the virtual freeform surface is obtained.

6. The background noise suppression system for microarray protein chips according to claim 5, characterized in that, Within each annular and radial region, using the fluorescence intensity distribution of the virtual freeform surface as a scalar field, the partial derivatives of the intensity values ​​in the horizontal and vertical coordinates are calculated pixel-by-pixel to construct a local gradient tensor matrix characterizing the spatial rate of change and directional features of background noise, including: Traverse the target pixels in each annular and radial region, and based on the grid topology of the virtual freeform surface, read the horizontal position index, vertical position index and corresponding fluorescence intensity fitting value of each target pixel in the two-dimensional pixel coordinate system. Using a preset discrete difference operator template, the center difference value of each target pixel in the horizontal and vertical coordinate axes is calculated respectively, and the horizontal and vertical partial derivative components that characterize the rate of change of background noise intensity along the horizontal and vertical axes are extracted. The horizontal and vertical partial derivative components corresponding to the same target pixel are orthogonally combined to construct a two-dimensional gradient vector containing information on the change magnitude and spatial gradient direction. The two-dimensional gradient vectors corresponding to all target pixels in each annular region and radial region are matrix-encapsulated in the order of their original physical coordinates to obtain a local gradient tensor matrix covering the entire target region.

7. The background noise suppression system for microarray protein chips according to claim 6, characterized in that, The normal offset correction coefficient is compared with the gray values ​​of corresponding pixels in the original fluorescence image using a point-by-point difference operation to obtain a preliminary background suppression image, including: The normal offset correction coefficient is received. Based on the physical coordinate system mapping relationship between the signal matrix to be processed and the virtual freeform surface mesh, the coordinates of each node of the normal offset correction coefficient matrix are spatially registered and aligned with the pixel coordinates of the signal matrix to be processed, and a one-to-one pixel-level mapping index is established. Based on pixel-level mapping index, the aligned pixel index positions are traversed, and the initial gray value of the corresponding pixel in the original fluorescence image and the normal offset correction coefficient matching the position are read synchronously to complete the data alignment before the difference operation. The normal offset correction coefficient and the corresponding initial gray value are subjected to point-by-point algebraic difference operation to remove the intensity superposition redundancy caused by spatial heterogeneous background noise, and a denoised intermediate gray value dataset carrying effective signal features is obtained. The negative gray-level outliers below the zero threshold generated by the difference operation in the denoised intermediate gray-level dataset are truncated and set to zero. The processed effective gray-level values ​​are then rearranged and encapsulated according to the original two-dimensional pixel grid to obtain a preliminary background suppression image.

8. The background noise suppression system for microarray protein chips according to claim 7, characterized in that, Anisotropic diffusion filtering is applied to the residual light scattering noise at the edges of each array point in the preliminary background-suppressed image to finally output a clean fluorescence signal image with suppressed background noise, including: Based on the virtual freeform surface, the Gaussian curvature values ​​of each grid node are extracted, and the Gaussian curvature values ​​are mapped to the pixel array of the signal matrix to be processed according to the physical coordinates to obtain the Gaussian curvature distribution parameter map that characterizes the spatial fluctuation law of background noise. Read the preliminary background suppression image, identify the physical boundaries of each array point through the edge contour extraction operator, extract the transition zone pixel region of a preset width outside the boundary, and construct a local noise mask that characterizes the spatial distribution of light scattering residual noise. The pixel region covered by the local noise mask is used as the diffusion iteration field. The anisotropic diffusion filtering algorithm is called to perform multiple rounds of pixel gray-level redistribution. In each iteration, the data at the corresponding position of the Gaussian curvature distribution parameter map is dynamically mapped to the adaptive regularization parameter of the diffusion control equation. The diffusion intensity and edge protection threshold are adjusted in real time according to the curvature gradient difference. When the diffusion iteration reaches the preset convergence condition or the pixel grayscale residual of adjacent iterations is lower than the preset threshold, the filtering operation is terminated, and the effective grayscale values ​​after iteration convergence are re-encapsulated according to the original two-dimensional pixel grid topology to obtain a pure fluorescence signal image with suppressed background noise.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to perform the system as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, performs the system as described in any one of claims 1 to 8.

Images