Microscopic image processing method and device and electronic equipment

Abstract

The invention discloses a microscopic image background removing method and device and electronic equipment, and belongs to the technical field of microscopic imaging. According to the method, on the basis of a traditional ball algorithm, global resolution, local gradient and intensity and edge gating three-dimensional features are introduced, a pixel-by-pixel adaptive radius matrix is generated through mapping and normalization, improved ball operation is guided by the pixel-by-pixel adaptive radius matrix, and high-precision, high-efficiency and robust background removal is achieved. Experiments show that the method is obviously superior to a fixed radius method on various microscopic images, does not need manual parameter adjustment, and is suitable for high-throughput automatic scenes.

Description

Technical Field

[0001] This invention relates to the field of digital image processing in microscopic imaging technology, specifically to a pixel-by-pixel adaptive rolling ball method, apparatus, and electronic device for background removal of microscopic images. It aims to solve the problems of insufficient accuracy, low efficiency, and poor robustness in background removal caused by complex backgrounds, variable structures, and noise interference in microscopic images. Background Technology

[0002] Microscopic imaging is widely used in life sciences and materials science, where background removal is a crucial preliminary step for subsequent quantitative analysis, 3D reconstruction, and artificial intelligence recognition. Among existing technologies, the Rolling Ball Algorithm (RoBAL) is widely adopted for background removal due to its simplicity and ease of implementation. This method constructs a spherical structural element with a fixed radius and "rolls" it across the image's grayscale surface, using the trajectory of the sphere's lowest point as the background surface. This surface is then subtracted from the original image to extract the foreground target. However, the traditional Rolling Ball Algorithm exhibits the following significant drawbacks in practical microscopic image processing:

[0003] First, the background removal accuracy is insufficient. Because microscopic images often exhibit uneven illumination, variations in sample thickness, and changes in the refractive index of the medium, a globally fixed-radius spherical structural element cannot adequately capture the local features of different regions within the image, leading to background residue or misidentification of the foreground. For example, in regions with dense cell nuclei, if the radius is too large, the sphere may span multiple cells, misidentifying the true foreground as background; conversely, in regions with flat backgrounds, if the radius is too small, it cannot effectively fit the slowly changing low-frequency background, leaving noticeable banded artifacts.

[0004] Secondly, it is highly parameter-dependent. Traditional methods require operators to repeatedly adjust the sphere radius manually based on experience, lacking objective standards. Images taken by different batches or different microscopes within the same batch need to have their radii calibrated individually, which is time-consuming and labor-intensive, making it difficult to achieve batch and automated processing.

[0005] Secondly, computational efficiency is low. When processing high-resolution (≥2048×2048) or ultra-large field-of-view microscopic images, fixing a large-radius sphere requires a large number of morphological operations across the entire image area. The time complexity increases with the image size and radius, for example, it can reach O(N·R). 2 The value is on the order of magnitude of 100 pixels, where N is the number of image pixels and R is the radius of the sphere. Experiments show that on a 3.2 GHz CPU platform, processing a single 4096×4096 microscopic fluorescence image with a radius of 100 pixels takes more than 45 seconds, which is insufficient to meet the requirements of high-throughput imaging.

[0006] Finally, it is sensitive to noise and fine structures. Microscopic images often contain a large amount of high-frequency noise due to photon scattering, detector readout noise, and sample autofluorescence. The fixed-radius rolling sphere algorithm lacks a noise preprocessing mechanism, and the sphere is easily "lifted" by sharp noise points during the rolling process, causing high-frequency jitter on the background surface, which in turn leads to the noise being misjudged as foreground signals. In addition, fine fibers, bubbles, or dust in the background can also interfere with the sphere's path, causing background estimation distortion.

[0007] In summary, existing rolling ball algorithms have significant shortcomings in terms of accuracy, efficiency, robustness, and automation in complex microscopic scenes, and there is an urgent need for a new background removal technique that can adaptively adjust the strategy based on the local content of the image.

[0008] Application content

[0009] The purpose of this application is to provide a method, apparatus, and electronic device for background removal of microscopic images. Through an improved rolling ball algorithm with pixel-wise adaptive radius, the following objectives are achieved: (1) eliminating background residue and foreground clipping caused by the traditional globally uniform radius, thus improving background removal accuracy; (2) establishing an adaptive parameter generation mechanism based on the global resolution and local grayscale features of the image, eliminating the need for manual parameter tuning; (3) optimizing the computation path, reducing the processing time for high-resolution microscopic images; and (4) enhancing the tolerance to noise and complex structures, thereby improving the robustness of the algorithm. The specific technical solutions and features of this application are as follows:

[0010] This method introduces three-dimensional features—"global resolution—local gradient—intensity / edge gating"—within the traditional rolling sphere framework. After mapping and normalization, a pixel-wise adaptive radius matrix is ​​generated, and rolling sphere background estimation with pixel-wise radius is performed accordingly, ultimately outputting a background-free microscopic image. The specific steps include:

[0011] S1. Image Acquisition and Optional Preprocessing:

[0012] (1) Obtain the original microscopic image I(x,y) by fluorescence microscopy, confocal microscopy or wide field microscopy, where x∈[0,W-1], y∈[0,H-1], and W and H are the image width and height;

[0013] (2) Normalize I to obtain I norm Used for feature calculation, while retaining the original copy for result output.

[0014] S2. Global resolution information extraction:

[0015] The curve is estimated using single-graph Fourier ring correlation (FRC), and the cutoff frequency is determined based on a threshold criterion (preferably 0.143) combined with curve smoothing / fitting. It is assumed that the algorithm internally generates two statistically independent spectral estimates, F1 and F2 (e.g., through internal splitting / bootstrapping), and their ring band correlation is...

[0016]

[0017] The cutoff frequency is obtained from FRC(k) = 0.143, and the resolution is calculated. The pixel resolution scale is then obtained by combining this with pixel spacing.

[0018]

[0019] S3. Local grayscale variation and edge gating extraction:

[0020] (1) The horizontal gradient G is calculated using the 3×3 Sobel operator. x With vertical gradient G y gradient magnitude

[0021]

[0022] Stabilize the lower bound clamp for low-texture regions:

[0023] DT(x,y)=max(DT(x,y),γ·Rpx),γ=0.23

[0024] (2) The corresponding L(x,y) is calculated using the 3×3 Laplacian operator, with a threshold τ. L =0.3 Constructing edge gating

[0025]

[0026] S4, Intensity-Gated Extraction:

[0027] (1) to I norm Perform OTSU automatic thresholding to obtain the threshold t*, then take...

[0028]

[0029] (2) Defining strength gating accordingly

[0030]

[0031] S5, Pixel-by-pixel adaptive radius estimation:

[0032] (1) Construct the weight matrix:

[0033] (2) Minimum value normalization:

[0034]

[0035] (3) Generate pixel-by-pixel radius and set upper limit: R adp (x,y)=clip(κ·Rpx ·W norm (x,y),0,R max )

[0036] Among them, R is preferred. max =40, κ=1. Implement General R adp Quantization is performed as an 8-bit unsigned integer to improve efficiency.

[0037] S6, Improved Rolling Ball Background Removal:

[0038] With R adp For a pixel-by-pixel radius map, a rolling sphere background estimation method that supports pixel-by-pixel radius maps is invoked, optionally including pre-smoothing. This yields background image B. bg Then, output the foreground:

[0039] I fg(x,y) =max(I(x,y)-B bg (x,y),0)

[0040] The foreground and background images are saved in 16-bit format.

[0041] Furthermore, it also includes an apparatus for performing the above method, the apparatus comprising:

[0042] (1) Microscopic image acquisition module: used to acquire raw microscopic images;

[0043] (2) Global resolution calculation module: Execute step S2;

[0044] (3) Local feature extraction module: Steps S3 and S4 are executed in parallel;

[0045] (4) Adaptive radius generation module: Execute step S5;

[0046] (5) Improve the ball handling module: Execute step S6;

[0047] (6) Result output module: Output I_fg and optionally save it as a 16-bit TIFF.

[0048] Furthermore, it also includes an electronic device that performs the above method, the electronic device including a processor, memory and a non-transient computer-readable medium storing the instructions of the above method, wherein the processor executes the instructions to implement steps S1-S6.

[0049] In summary, this application addresses the following issues: Firstly, it dynamically generates a radius matrix (S2-S5) that matches local grayscale changes and background complexity, replacing the traditional fixed radius and achieving precise modeling that is "large when it should be large, and small when it should be small." Secondly, it addresses the issue of strong parameter dependence by fully binding radius calculation to image content, using FRC, gradient, and OTSU results as objective quantitative indicators, eliminating the need for manual intervention. Thirdly, it addresses the issue of low computational efficiency by utilizing the sparse distribution characteristics of the radius matrix and combining it with GPU parallelism to significantly reduce redundant computations. Fourthly, it addresses the issue of noise sensitivity by employing multiple denoising mechanisms, including optional smoothing in S1, gradient stabilization and gating in S3, and logarithmic mapping and minimum value normalization in S5, effectively suppressing noise interference on the rolling ball path. The specific beneficial effects achieved by this application are as follows:

[0050] 1. Improved accuracy: The pixel-by-pixel adaptive radius makes the background surface more closely match the real low-frequency changes. Experiments show that the mean square error of the foreground is reduced by 42%-57% compared with the traditional method on various samples such as cells, tissue sections, and microspheres.

[0051] 2. Automation: No manual radius setting is required throughout the process, reducing reliance on operator experience and making it suitable for unattended high-throughput scenarios.

[0052] 3. Efficiency optimization: Feature-driven radius distribution avoids large-radius calculations in flat areas, reducing the overall computational load by 30%-50%.

[0053] 4. Enhanced robustness: By fusing gradient and background priors and suppressing noise interference, the background can still be extracted stably even under extreme conditions where the signal-to-noise ratio is as low as 10dB. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the overall process of the image background removal method described in this application. Detailed Implementation

[0055] The term "embodiment" used herein, as an example, is not necessarily to be construed as superior to or better than other embodiments. Performance testing in the embodiments of this application, unless otherwise specified, employs conventional testing methods in the art. It should be understood that the terminology used in this application is merely for describing particular implementations and is not intended to limit the scope of this disclosure.

[0056] Unless otherwise stated, the technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; other experimental methods and technical means not specifically mentioned herein refer to experimental methods and technical means commonly used by one of ordinary skill in the art.

[0057] To better illustrate the content of this application, numerous specific details are provided in the following detailed embodiments. Those skilled in the art should understand that this application can be implemented even without certain specific details. In the embodiments, some methods, means, instruments, and devices well-known to those skilled in the art are not described in detail in order to highlight the main points of this application.

[0058] The specific embodiments of this application are described below to enable those skilled in the art to understand this application. However, it should be understood that this application is not limited to the scope of the specific embodiments. For those skilled in the art, as long as the various changes are within the spirit and scope of this application as defined and determined by the appended claims, these changes are obvious. All applications that utilize the concept of this application are protected.

[0059] Example

[0060] Please see Figure 1 The flowchart illustrates the image processing method, apparatus, and electronic device for background removal in microscopic images proposed in this application. Two embodiments are provided, and the implementation process of the technical solution in actual production is detailed below through specific embodiments. The specific implementation methods are as follows:

[0061] Example 1

[0062] Taking background removal from fluorescence wide-field microscopy images as an example, the process involves S1: image acquisition and optional preprocessing. Image parameters: 2048×2048 pixels, 16-bit grayscale, objective lens 40×, NA 0.95. The original image is linearly normalized to obtain I0 for feature calculation. norm .

[0063] Perform S2, extract global resolution information, and obtain R. px =75.76px, pixel size 30.6nm, corresponding resolution d=2.32μm.

[0064] Execute steps S3-S5 to generate the R_adp matrix, with a mean of 38 pixels and a variance of 8 pixels.

[0065] The processing time was 1.8 seconds. After background removal, the foreground signal-to-noise ratio was improved by 6.3 dB, and the cell nucleus segmentation F1-score was improved from 0.81 to 0.93.

[0066] Example 2

[0067] Taking the confocal Z-stack maximum projection image as an example. Perform S1, image acquisition and optional preprocessing: Image parameters: 4096×4096 pixels, Z-stack 50 layers, maximum projection; perform linear normalization on the original image to obtain I for feature calculation. norm .

[0068] Perform S2, extract global resolution information, and obtain R. px =102.04px, pixel size 61.2nm, corresponding resolution d=6.24μm.

[0069] Executing S3-S5 generates the R_adp matrix, performs CUDA parallel computation, GPU RTX 3080, takes 3.4 seconds.

[0070] After background suppression, the vascular network continuity index increased by 22%, and subsequent 3D reconstruction artifacts were significantly reduced.

[0071] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0072] The technical solutions and technical details disclosed in the embodiments of this application are merely illustrative of the application concept and do not constitute a limitation on the technical solutions of this application. Any conventional changes, substitutions or combinations made to the technical details disclosed in the embodiments of this application have the same application concept as this application and are within the protection scope of the claims of this application.

Claims

1. A method for removing background from microscopic images, characterized in that, Includes the following steps: S1. Obtain the original microscopic image and perform linear normalization for subsequent processing; S2. Based on the single-image Fourier ring correlation method (FRC), a threshold criterion is used in combination with curve fitting to obtain the scale parameter reflecting the spatial resolution of the image, which is then used for subsequent radius estimation. S3. Extract the local grayscale change matrix of the image using a differential operator, and generate an edge activity gating mask based on a preset threshold; S4. Generate an intensity-gated mask using the OTSU automatic threshold segmentation algorithm; S5. Combining the resolution scale of step S2, the local grayscale change matrix of step S3, and the two types of gated masks mentioned in S3 and S4, a pixel-wise adaptive radius matrix R_adp is generated after mapping and normalization. S6. Using R_adp as the pixel-by-pixel radius and setting a global radius upper limit, execute the improved rolling ball background removal algorithm, including pre-smoothing, to obtain the background image B_bg; S7. Subtract B_bg from the original image and output the microscopic image after removing the background.

2. The method according to claim 1, wherein step S2 involves performing a single-image Fourier ring correlation (FRC) analysis on the image, determining the cutoff frequency based on a threshold criterion and in conjunction with curve smoothing or fitting, and thereby obtaining a resolution scale parameter for radius estimation, with the threshold set to 0.

143.

3. The method according to claim 1, wherein step S3 comprises: (1) Differential operators are used to extract the degree of local gray-level changes, with the 3×3 Sobel operator being preferred; (2) An edge activity gating mask is generated based on the threshold of the differential response using a 3×3 Laplacian operator; (3) Stabilize low-texture areas.

4. The method according to claim 1, wherein in step S5, based on the resolution scale of step S2, the local grayscale change of step S3, and the gated mask of steps S3 and S4, a pixel-wise adaptive radius matrix is ​​generated by combining mapping and normalization, and a preset upper limit is set for the radius value, and the radius matrix is ​​quantized into an 8-bit integer to improve computational efficiency.

5. The method according to claim 1, wherein step S6 comprises: Using the pixel-by-pixel radius matrix as input and setting a global radius upper limit, an improved rolling ball background removal algorithm is executed to estimate the background image, and pre-smoothing is performed before rolling ball processing.

6. A microscopic image background removal apparatus, comprising: (1) Microscopic image acquisition module; (2) Global resolution calculation module; (3) Local feature extraction module; (4) Adaptive radius generation module; (5) Improve the ball handling module; (6) Results output module. The improved rolling ball processing module is configured to receive a pixel-by-pixel radius matrix and perform background estimation in conjunction with a global radius upper limit, and optionally includes pre-smoothing. The apparatus is used to implement the method according to any one of claims 1-5.

7. The apparatus of claim 6, wherein the improved rolling ball processing module is optionally implemented in parallel, including but not limited to multithreading or GPU acceleration.

8. An electronic device comprising a processor, a memory, and a computer program stored in the memory, wherein the processor, when executing the program, implements the method of any one of claims 1-5.

9. A non-transient computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the method of any one of claims 1-5.

Images