Protein point identification and analysis method and device based on real-time cell imaging

Abstract

The invention belongs to the technical field of cytobiology and optical imaging, and particularly relates to a protein point recognition and analysis method and device based on real-time cell imaging. The protein point recognition method based on real-time cell imaging comprises the following steps: S11, acquiring a cell fluorescence image, performing denoising processing on the cell fluorescence image by adopting a non-local mean denoising algorithm, and removing background nonuniformity by using morphological opening operation to obtain a preprocessed image; and S12, dynamically calculating a threshold value of the preprocessed image based on local image intensity, carrying out binaryzation to obtain an image with a background and a foreground, carrying out morphological operation on the image to remove noise and artifacts, and carrying out image region segmentation on the image to obtain an image with a plurality of protein points. According to the method, a non-local mean denoising algorithm and morphological background correction are combined, so that the quality of the low-signal-to-noise-ratio cell fluorescence image is remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of cell biology and optical imaging technology, and particularly relates to a protein spot recognition and analysis method and device based on real-time cell imaging. Background Art

[0002] In the prior art, the detection and analysis of protein spots in cell images are usually completed using analysis software or analysis tools.

[0003] 1. Image processing-based protein spot detection and analysis software (Journal Name: Bioinformatics, Volume Number: 36, Issue Number: 10, Page Number: 3215-3222, Publication Year: 2020) uses image processing technology to automatically detect and quantify protein spots in cell fluorescence images. Its core algorithms include image preprocessing (such as denoising and background correction), protein spot detection (based on threshold segmentation or edge detection), and fluorescence intensity quantification. The software also provides simple data visualization functions, such as protein spot distribution heat maps and fluorescence intensity curves. Its shortcomings include:

[0004] (1) Poor detection effect on low signal-to-noise ratio images: The software mainly relies on threshold segmentation and edge detection algorithms, which have high requirements on image quality. When the image signal-to-noise ratio is low, noise can be easily misidentified as protein spots, resulting in inaccurate detection results.

[0005] Reason: Traditional image processing algorithms lack the ability to adapt to complex backgrounds and have difficulty distinguishing between noise and real protein spots.

[0006] (2) Lack of dynamic tracking capability: The software only supports static analysis of a single image and cannot track the dynamic changes of protein spots (such as changes in fluorescence intensity over time) in real time.

[0007] Reason: The software design did not take into account the processing of time series data and lacked the function of correlation analysis of multiple frame images.

[0008] (3) Simple data analysis function: The software only provides basic fluorescence intensity statistics and distribution heat maps, and cannot support complex multi-parameter statistical analysis (such as protein spot shape factor, cluster analysis, etc.).

[0009] Reason: The software function design is relatively simple and does not integrate advanced data analysis tools.

[0010] 2. A machine learning-based protein spot quantification analysis tool (Patent Document Number: US2019 / 0156789A1, Patent Name: Protein Spot Detection and Quantification System Using Machine Learning, Publication Date: May 23, 2019) uses machine learning algorithms (such as convolutional neural networks, CNNs) to automatically identify and quantify protein spots in cell fluorescence images. By training a deep learning model, it can accurately identify protein spots from complex backgrounds and quantify their fluorescence intensity, shape, and distribution. The tool also supports multi-image batch processing and big data analysis. Its shortcomings include:

[0011] (1) Dependence on a large amount of annotated data: This tool uses a deep learning model for protein spot detection, which requires a large amount of annotated data for model training. However, obtaining annotated data is time-consuming and costly, limiting the tool's universality.

[0012] Reason: The performance of deep learning models is highly dependent on the quality and quantity of training data, while protein spot annotation requires professional knowledge and a lot of manual input.

[0013] (2) High computing resource requirements: The training and inference processes of deep learning models require high hardware computing resources (such as GPUs), and ordinary computers are difficult to meet their operating requirements.

[0014] Reason: Deep learning algorithms have high computational complexity, especially when processing high-resolution cell images.

[0015] (3) Lack of real-time analysis capability: This tool is mainly used for offline analysis and cannot track and analyze the dynamic changes of protein spots in real time.

[0016] Reason: The reasoning speed of deep learning models is slow, which makes it difficult to meet the needs of real-time data processing.

[0017] (4) Limited model generalization ability: The trained model has poor generalization ability under different experimental conditions or cell types, which may lead to inaccurate detection results.

[0018] Reason: Deep learning models are sensitive to the distribution of training data. When the input data differs significantly from the training data, the model performance will drop significantly. Summary of the Invention

[0019] The present invention addresses the technical problems in the prior art of protein spot detection and analysis, such as high image quality requirements, lack of dynamic analysis capabilities, single functions or high computational costs. The purpose of the present invention is to provide a protein spot identification and analysis method and device based on real-time cell imaging.

[0020] In order to solve the aforementioned technical problems, the first aspect of the present invention provides a protein spot identification method based on real-time cell imaging, the protein spot identification method based on real-time cell imaging comprising:

[0021] S11, acquiring a cell fluorescence image, performing denoising on the cell fluorescence image using a non-local means denoising algorithm, and using a morphological opening operation to remove background inhomogeneity to obtain a preprocessed image;

[0022] S12, dynamically calculating a threshold of the preprocessed image based on the local image intensity and binarizing it to obtain an image with a background and a foreground, removing noise and artifacts from the image through morphological operations, and segmenting the image to obtain an image with several protein spots.

[0023] Optionally, in the protein spot identification method based on real-time cell imaging as described above, in step S11, the cell fluorescence image is denoised using a non-local means denoising algorithm, which is implemented using the following formula:

[0024]

[0025] Among them, x represents the position of the current pixel to be denoised, I denoised (x) represents the intensity value of pixel x in the denoised image, C(x) represents the normalization constant, y represents the position of another pixel in the image, Ω represents the set of all pixels y in the image that may participate in the calculation, w(x, y) is the weight function, which represents the contribution of pixel y to the denoising of pixel x, and I(y) represents the intensity value of pixel y in the original image.

[0026] Optionally, in the protein spot recognition method based on real-time cell imaging as described above, in step S11, a morphological opening operation is used to remove background inhomogeneity, which is implemented using the following formula:

[0027] I corrected =I original –Opening(I original )

[0028] Among them, I corrected represents the image after background correction, I original Represents the original input image, Opening(I original ) represents the result of morphological opening operation on the original input image.

[0029] Optionally, in the protein spot identification method based on real-time cell imaging as described above, in step S12, the threshold of the pre-processed image is dynamically calculated based on the local image intensity and binarized, which is achieved using the following formula:

[0030] T(x)=μ(x)+k·σ(x)

[0031] Where T(x) represents the dynamic threshold of pixel x, μ(x) represents the average grayscale value of the pixel x, σ(x) represents the standard deviation of the grayscale value of the pixel x, and k is the adjustment coefficient.

[0032] According to the T(x) result, the pixels are divided into foreground and background categories to generate a binary image.

[0033] Optionally, in the protein spot identification method based on real-time cell imaging as described above, in step S12, when removing noise and artifacts from the image, it is achieved through a morphological opening operation or a closing operation.

[0034] In order to solve the aforementioned technical problems, the second aspect of the present invention provides a protein spot analysis method based on real-time cell imaging, the protein spot analysis method based on real-time cell imaging comprising:

[0035] S21, based on the Hungarian algorithm, matches protein spots in multiple frames of images obtained by the protein spot recognition method based on real-time cell imaging, and calculates the dynamic parameters of protein spots in real time;

[0036] S22, calculating the shape factor of the detected protein spots, clustering the protein spots in the image using a clustering analysis algorithm, and marking one or more types of protein spots;

[0037] S23, the analysis results including the dynamic parameters, shape factors and types of the protein spots are visualized and an analysis report is generated.

[0038] Optionally, in the protein spot analysis method based on real-time cell imaging as described above, in step S21, the protein spots in the multi-frame images obtained by the protein spot recognition method based on real-time cell imaging are matched based on the Hungarian algorithm, which is implemented using the following formula:

[0039] Cost Matrix ij =||Position i -Position j ||+λ||Intensity i -Intensity j ||

[0040] Among them, Cost Matrix ij represents the matching cost between the i-th protein point and the j-th protein point, ||Position i -Position j|| represents the position difference between the i-th protein spot and the j-th protein spot, λ represents the weight coefficient, and ||Intensity i -Intensity j || represents the intensity difference between the i-th protein spot and the j-th protein spot.

[0041] Optionally, in the protein spot analysis method based on real-time cell imaging as described above, in step S21, the dynamic parameters of the protein spots include at least one of a change in the number of protein spots, a change in the fluorescence intensity of the protein spots, a change in the area of ​​the protein spots, and a change in the distribution of the protein spots;

[0042] Real-time calculation of the number changes of protein spots is achieved by the following method:

[0043] Assume that the sets of protein points in two adjacent frames of images are {P1, P2, ..., PN} and {Q1, Q2, ..., QM} respectively;

[0044] The Hungarian algorithm is used to determine the number of protein spots that are successfully matched between two adjacent frames of images as K;

[0045] The following quantity change calculation is performed on the protein spots in two adjacent frames of images:

[0046] New protein point = MK

[0047] Disappearing protein spots = NK

[0048] ΔN=MN

[0049] ΔN represents the net change of protein spots in two adjacent frames. When ΔN>0, it means the number of protein spots increases; when ΔN<0, it means the number of protein spots decreases; when ΔN=0, it means the number of protein spots remains unchanged.

[0050] Real-time calculation of the fluorescence intensity changes of protein spots is achieved by the following method:

[0051] Assume that the protein points that are successfully matched in two adjacent frames of images are (Pi, Qj) determined by the Hungarian algorithm;

[0052] Calculate the intensity difference ΔI for the successfully matched protein spots ij :

[0053] ΔI ij =Intensity(Q j )-Intensity(P i )

[0054] Among them, Intensity (Q j ) represents protein spot Q j Intensity (P i) represents protein dot P i Strength;

[0055] Calculate the mean intensity change:

[0056]

[0057] Where K represents the number of protein point pairs that are successfully matched between two adjacent image frames determined by the Hungarian algorithm;

[0058] Real-time calculation of the area change of protein spots is achieved by the following method:

[0059] Assume that the protein points that are successfully matched in two adjacent frames of images are (Pi, Qj) determined by the Hungarian algorithm;

[0060] Calculate the pixel area of ​​each protein spot;

[0061] Calculate the area change ΔA for the successfully matched protein spots ij :

[0062] ΔA ij =Area(Q j )-Area(P i )

[0063] Among them, Area(Q j ) represents protein spot Q j The pixel area, Area(P i ) represents protein dot P i The pixel area;

[0064] Calculate the average area change:

[0065]

[0066] Where K represents the number of protein point pairs that are successfully matched between two adjacent image frames determined by the Hungarian algorithm;

[0067] Calculate the area expansion / contraction ratio:

[0068] Expansion Ratio=Area(Q j ) / Area(P i )

[0069] Real-time calculation of protein spot distribution changes is achieved by:

[0070] Calculate the moving distance ΔD for the successfully matched protein spots ij :

[0071] ΔD ij =||Position(D j )-Position(Di )||

[0072] Among them, Position(D j ) indicates that the protein points D that are successfully matched in two adjacent frames are determined by the Hungarian algorithm. j The center of mass position, Position(D i ) indicates that the protein points D that are successfully matched in two adjacent frames are determined by the Hungarian algorithm. i The center of mass position of ||Position(Qj)-Position(Pi)|| represents the protein point D that is successfully matched j Protein D i the distance between them;

[0073] Calculate the mean displacement:

[0074]

[0075] Here, K represents the number of protein point pairs that are successfully matched in two adjacent image frames determined by the Hungarian algorithm.

[0076] Alternatively, in the protein spot analysis method based on real-time cell imaging as described above, the intensity difference ΔI is calculated for the successfully matched protein spots. ij Finally, the intensity changes of individual protein spots are displayed through histograms or box plots.

[0077] Alternatively, in the protein spot analysis method based on real-time cell imaging as described above, the moving distance ΔD of the successfully matched protein spots is calculated. ij Finally, the migration direction of the protein spots was determined.

[0078] Optionally, in the protein spot analysis method based on real-time cell imaging as described above, after the distribution changes of the protein spots are calculated in real time, the spatial distribution differences between the two frames are compared by kernel density estimation.

[0079] Optionally, in the protein spot analysis method based on real-time cell imaging as described above, in step S22, the shape factor of the detected protein spots is calculated using the following formula:

[0080]

[0081] Among them, Circularity represents the circularity, Area represents the area of ​​the protein dot, and Perimeter represents the perimeter of the protein dot.

[0082] Optionally, in the protein spot analysis method based on real-time cell imaging as described above, in step S22, the cluster analysis algorithm adopts the DBSCAN algorithm.

[0083] Optionally, in the protein spot analysis method based on real-time cell imaging as described above, in step S23, when visually displaying the analysis results, a preset visualization tool is used to present the analysis results in the form of a chart;

[0084] In step S23, the analysis report includes analysis data and charts, and supports export operations in several different formats.

[0085] In order to solve the aforementioned technical problems, the third aspect of the present invention provides a protein spot recognition and analysis device based on real-time cell imaging, the protein spot recognition and analysis device based on real-time cell imaging comprising:

[0086] An image preprocessing module is used to obtain a cell fluorescence image, perform denoising on the cell fluorescence image using a non-local means denoising algorithm, and use a morphological opening operation to remove background inhomogeneity to obtain a preprocessed image;

[0087] The protein spot detection module is used to dynamically calculate the threshold of the preprocessed image based on the local image intensity and perform binarization to obtain an image with background and foreground. The image is subjected to morphological operations to remove noise and artifacts, and the image is segmented to obtain an image with several protein spots.

[0088] Real-time data analysis module, used to match protein spots in multiple frames of images based on the Hungarian algorithm and calculate the dynamic parameters of protein spots in real time;

[0089] Multi-parameter statistical analysis module, used to calculate the shape factor of the detected protein spots, cluster the protein spots in the image using a clustering analysis algorithm, and label one or more types of protein spots;

[0090] The data visualization and report generation module is used to visualize the analysis results, including the dynamic parameters, shape factors, and types of protein spots, and generate analysis reports.

[0091] The positive progress effect of the present invention is:

[0092] 1. Optimize image preprocessing algorithm: The present invention combines the non-local means denoising algorithm (NLM) with morphological background correction to significantly improve the quality of low signal-to-noise ratio cell fluorescence images.

[0093] Specifically, the NLM algorithm can effectively remove random noise while preserving protein spot details. Morphological background correction solves the problem of uneven image background, providing a clear data foundation for subsequent protein spot detection.

[0094] 2. Adaptive protein spot detection: The present invention uses dynamic threshold segmentation and morphological filtering to improve the accuracy and robustness of protein spot detection.

[0095] Specifically, the present invention proposes a dynamic threshold segmentation algorithm that dynamically calculates thresholds based on local image intensity, adapting to the needs of protein spot detection in different regions. Adaptive threshold segmentation addresses the poor performance of traditional fixed threshold methods in complex images. Morphological filtering (such as opening and closing operations) removes noise and artifacts, further optimizing detection results, improving accuracy, and reducing false positives and missed detections.

[0096] 3. Real-time dynamic tracking and analysis: Protein spot matching technology based on the Hungarian algorithm enables dynamic tracking of protein spots in multiple frames. The Hungarian algorithm can efficiently and accurately match protein spots in multiple frames, solving the association problem in dynamic tracking.

[0097] Real-time calculation of protein spot number, fluorescence intensity, area, and distribution changes supports time series data analysis. This real-time data analysis capability fills a gap in existing technologies for studying dynamic protein spot behavior.

[0098] 4. Multi-parameter statistical analysis function: integrated shape factor calculation and cluster analysis, providing multi-dimensional protein spot data analysis.

[0099] Specifically, shape factor calculation can quantify the morphological characteristics of protein spots, providing more information for cell biology research. Cluster analysis reveals the spatial distribution patterns of protein spots, supporting deeper biological interpretations.

[0100] 5. User-friendly interface design and report generation: Provides intuitive data visualization tools (such as protein spot distribution heat map, fluorescence intensity curve) and automatic report generation function.

[0101] Specifically, through modular design and intuitive visualization tools, the user interface is simple and easy to use, lowering the barrier to entry for non-expert users. Automated report generation significantly improves research efficiency and supports export to various formats such as PDF and Excel. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] The disclosure of the present invention will become more apparent with reference to the accompanying drawings. It should be understood that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. In the drawings:

[0103] Figure 1 A flow chart of the present invention;

[0104] Figure 2 A schematic diagram of a cell fluorescence image of the present invention;

[0105] Figure 3 For Figure 2 A schematic diagram after being processed by the present invention;

[0106] Figure 4 A data analysis display diagram of the present invention;

[0107] Figure 5 Another data analysis display diagram of the present invention;

[0108] Figure 6 The figure is an application flow chart of the present invention. DETAILED DESCRIPTION

[0109] The following describes the embodiments of the present invention through specific examples. Those skilled in the art will readily understand the other advantages and benefits of the present invention from the disclosure herein. The present invention may also be implemented or applied through various other specific embodiments, and the details in this specification may be modified or altered based on different perspectives and applications without departing from the spirit of the present invention.

[0110] It should be noted that, unless there is any conflict, the following embodiments and features therein may be combined with each other.

[0111] In the description of the present invention, it should be noted that, for directional words, such as the terms "outside", "middle", "inside", "outside", etc., the directions and positional relationships indicated are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and cannot be understood as limiting the specific scope of protection of the present invention.

[0112] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features. Therefore, the terms "first" and "second" may explicitly or implicitly refer to one or more of these features. Throughout the description of the present invention, "several" and "a number" mean two or more, unless otherwise specifically defined.

[0113] Reference Figure 1 The embodiment of the present invention provides a protein spot recognition method based on real-time cell imaging, which includes the following specific steps:

[0114] S11, image preprocessing: acquiring a cell fluorescence image, performing denoising on the cell fluorescence image using a non-local means denoising algorithm, and using a morphological opening operation to remove background inhomogeneity to obtain a preprocessed image.

[0115] S12, protein spot detection: The preprocessed image is thresholded dynamically based on the local image intensity and binarized to obtain an image with background and foreground. The image is subjected to morphological operations to remove noise and artifacts, and the image is segmented to obtain an image with several protein spots.

[0116] The cell fluorescence images obtained by the present invention are as follows Figure 2 As shown in FIG, step S11 pre-processes the input cell fluorescence image by denoising and background correction to improve the image quality and provide clear data for subsequent protein spot detection. Step S12 automatically identifies protein spots from the pre-processed image by adaptive threshold segmentation, morphological filtering and image region segmentation. Figure 3 As shown, Figure 3 The background in is the state after processing in step S11. After processing in step S12, Figure 3 The green color in the figure represents the lane containing protein spots.

[0117] Prior art methods typically use image processing algorithms and / or deep learning models to process cell fluorescence images. This can easily misidentify noise as protein spots in images with low signal-to-noise ratios, leading to inaccurate detection results. This invention significantly improves the accuracy of protein spot recognition in low-SNR images by optimizing image preprocessing algorithms and introducing adaptive detection technology.

[0118] This invention abandons deep learning tools and adopts a lightweight algorithm design, which reduces the computing resource requirements while ensuring detection accuracy, enabling it to run efficiently on ordinary computers.

[0119] In some embodiments, in step S11, the cell fluorescence image is denoised using a non-local means denoising algorithm (NLM) to remove random noise in the image, which is achieved using the following formula:

[0120]

[0121] Where x represents the position of the pixel to be denoised (usually a two-dimensional coordinate, such as x = (i, j)); I denoised(x) represents the intensity value of pixel x in the denoising image (i.e., the denoising result); C(x) represents the normalization constant, which ensures that the sum of the weights is 1; y is the position of another pixel in the image (similar to x, a two-dimensional coordinate), which is used to calculate the similarity weight with x; Ω represents the set of all pixels y in the image that may participate in the calculation (usually the entire image, but preferably limited to a preset search window); w(x, y) is a weight function, which represents the contribution of pixel y to the denoising of pixel x, and the weight is determined by the similarity between the neighborhood of pixel x and pixel y; I(y) represents the intensity value of pixel y in the original image (i.e., the pixel value of the input image).

[0122] In this embodiment, the intensity value is the grayscale value / pixel value of the image. Therefore, if the cell fluorescence image obtained in step S11 is not a grayscale image, it is first grayscale processed to obtain a grayscale image, so that the information in the grayscale image can be extracted. However, the grayscale image will not be displayed on the interactive interface for the user to view. The display interface usually displays the color image after image enhancement, that is, pre-processing. In a grayscale image, the grayscale value of each pixel is usually represented by a numerical value, ranging from pure black to pure white. Therefore, the grayscale value is a quantitative representation of the brightness of the pixel, reflecting the brightness and darkness of the point. In an 8-bit grayscale image, the grayscale value range is 0 to 255, where 0 represents pure black (darkest), 255 represents pure white (brightest), and intermediate values ​​(such as 128) represent different shades of gray. In grayscale images with higher bit depth (such as 16 bits), the grayscale value range may be larger (such as 0 to 65535), providing finer brightness levels.

[0123] In some embodiments, in step S11, a morphological opening operation is used to remove background non-uniformity, which is implemented using the following formula:

[0124] I corrected =I original –Opening(I original )

[0125] Among them, I corrected Represents the image after background correction; I original Represents the original input image, that is, the image after denoising, usually a grayscale image; Opening(I original ) represents the result of morphological opening operation (erosion followed by dilation) on the original input image.

[0126] Even after denoising cell fluorescence images using the non-local means denoising algorithm, uneven illumination, background gradients, or noise can lead to inconsistent contrast between the target object and the background, compromising subsequent analysis. Therefore, in this real-time example, an opening operation is used to remove bright areas smaller than the structuring element (such as noise or small objects) in the image while preserving the large-scale background structure.

[0127] Therefore, Opening(I original ) can be viewed as an estimation of the background in the original input image (approximate illumination or background distribution).

[0128] The calculation logic of this real-time example is: by subtracting the estimated background (the result of the opening operation) from the original input image, the true signal of the target object is retained while eliminating background non-uniformity. In other words, the background-corrected image = original image - estimated background. The corrected image = original image - estimated background.

[0129] If the background of a certain area in the original input image is brighter (such as a bright spot with uneven lighting), the opening operation in this embodiment will retain the bright spot, and after subtraction, the area will be darkened to achieve uniformity. Therefore, the contrast of the target object (which is usually brighter or darker than the background) will be enhanced.

[0130] In some embodiments, in step S12, the pre-processed image is thresholded dynamically based on the local image intensity and binarized using the following formula:

[0131] T(x)=μ(x)+k·σ(x)

[0132] Where T(x) represents the dynamic threshold of pixel x; μ(x) represents the average grayscale value of the pixel x; σ(x) represents the standard deviation of the grayscale value of the pixel x; k is the adjustment coefficient used to control the sensitivity of segmentation.

[0133] According to the T(x) result, the pixels are divided into foreground and background categories to generate a binary image.

[0134] This embodiment adopts adaptive threshold segmentation based on the Niblack binary segmentation algorithm, which can adaptively select thresholds according to the brightness of different areas, thereby effectively separating the foreground and background.

[0135] The Niblack binary segmentation algorithm calculates a dynamic threshold based on the local grayscale value of each pixel, and divides the pixels into foreground and background according to the dynamic threshold.

[0136] Because the Niblack binary segmentation algorithm is sensitive to noise and has poor adaptability to sudden changes in image brightness, prior to using this embodiment, the input cell fluorescence image is effectively preprocessed using denoising and background correction in step S11 to overcome these shortcomings of the Niblack binary segmentation algorithm.

[0137] In some embodiments, in step S12, when removing noise and artifacts from the image, it is achieved through a morphological opening operation or a closing operation.

[0138] In some embodiments, in step S12 , when performing image region segmentation on the image, an existing image region segmentation algorithm, such as watershed transform, is used.

[0139] The protein spot recognition method based on real-time cell imaging of the present invention can realize protein spot detection by processing the real-time input cell fluorescence image through steps S11 and S12.

[0140] Reference Figure 1 The present invention also provides a protein spot analysis method based on real-time cell imaging. The protein spot analysis method based on real-time cell imaging includes the following specific steps:

[0141] S21, real-time data analysis: Using the Hungarian algorithm, protein spots are matched in multiple frames of images obtained by the protein spot recognition method based on real-time cell imaging, and the dynamic parameters of the protein spots are calculated in real time;

[0142] S22, multi-parameter statistical analysis: shape factor calculation of detected protein spots, clustering of protein spots in the image using clustering analysis algorithm, and labeling of one or more types of protein spots;

[0143] S23, Data Visualization and Report Generation: Visualize the analysis results including the dynamic parameters, shape factors, and types of protein spots and generate an analysis report.

[0144] Step S21 dynamically tracks and quantifies protein spots in the time series images using a protein spot matching algorithm and dynamic parameter calculation. Step S22 performs multi-dimensional statistical analysis of the detected protein spots using shape factor calculation and cluster analysis. Step S23 presents the analysis results in the form of charts and reports using visualization tools and analysis reports.

[0145] Existing technologies primarily focus on static image analysis and lack the ability to track the dynamic behavior of protein spots (e.g., changes in fluorescence intensity over time) in real time. This invention incorporates real-time data analysis, enabling continuous analysis of time-series images and dynamic tracking of protein spot parameters (e.g., number of spots, fluorescence intensity, and distribution).

[0146] Existing technologies offer limited analytical capabilities, supporting only basic fluorescence intensity statistics and distribution heatmaps, and are unable to meet complex research needs. This new technology integrates multi-parameter statistical analysis, supports advanced functions such as protein spot area, shape factor, and cluster analysis, and provides an intuitive data visualization interface.

[0147] The deep learning models used in existing technologies have limited generalization capabilities under different experimental conditions and are complex to operate, making them difficult for non-expert users to accept. This invention, through modular design and a user-friendly interface, lowers the barrier to use while enhancing the algorithm's adaptability to different experimental conditions.

[0148] In some embodiments, in step S21, protein spots in multiple frames of images obtained by a protein spot recognition method based on real-time cell imaging are matched based on the Hungarian Algorithm to achieve matching of protein spots in multiple frames of images, which is achieved using the following formula:

[0149] Cost Matrix ij =||Position i -Position j ||+λ||Intensity i -Intensity j ||

[0150] Among them, Cost Matrix ij Represents the matching cost between the i-th protein point and the j-th protein point; ||Position i -Position j || represents the position difference between the i-th protein spot and the j-th protein spot (that is, the Euclidean distance); λ represents the weight coefficient, which is used to balance the influence of position and intensity on the matching; ||Intensity i -Intensity j || represents the intensity (grayscale value) difference (absolute difference) between the i-th protein spot and the j-th protein spot.

[0151] This embodiment calculates the cost matrix using the above formula, and calculates the best match from the cost matrix according to the Hungarian algorithm, that is, the target (i.e., protein point) between consecutive frames has the minimum cost to match the target pair (i.e., protein point pair) between two adjacent frames.

[0152] In some embodiments, in step S21 , the dynamic parameters of the protein spots include at least one of a change in the number of protein spots, a change in the fluorescence intensity of the protein spots, a change in the area of ​​the protein spots, and a change in the distribution of the protein spots.

[0153] One, multiple or all of the above dynamic parameters can be selected for calculation according to actual needs.

[0154] (1) Real-time calculation of the number change of protein spots (Number Change) is achieved by the following method:

[0155] (11) Input

[0156] The protein point sets {P1, P2, ..., PN} and {Q1, Q2, ..., QM} of two adjacent frame images (Frame1 and Frame2).

[0157] Therefore, let the sets of protein points in two adjacent frames of images be {P1, P2, ..., PN} and {Q1, Q2, ..., QM} respectively.

[0158] (12) Matching results

[0159] Matched points: protein points successfully paired by the Hungarian algorithm.

[0160] Unmatched points: points that appear in Frame 2 (new additions, possibly indicating newly expressed proteins) or points that disappear in Frame 1 (decreased, possibly indicating protein degradation or migration out of the field of view).

[0161] (13) Calculation formula

[0162] New protein point = MK

[0163] Disappearing protein spots = NK

[0164] ΔN=MN

[0165] Among them, the Hungarian algorithm is used to determine that the number of successfully matched protein spots in two adjacent image frames is K; ΔN represents the net change of protein spots in two adjacent image frames. When ΔN>0, it means that the number of protein spots increases, when ΔN<0, it means that the number of protein spots decreases, and when ΔN=0, it means that the number of protein spots remains unchanged.

[0166] (14) Example

[0167] Frame1 has 100 protein spots and Frame2 has 110 protein spots, of which 90 protein spots are successfully matched.

[0168] New protein points = 110-90 = 20

[0169] Disappearance of protein point = 100-90 = 10

[0170] Net change ΔN = 110 - 100 = +10.

[0171] (2) Real-time calculation of the fluorescence intensity change of the protein spot (Intensity Change) is achieved by the following method:

[0172] (21) Matching point pairs

[0173] Calculate the intensity difference ΔI for the protein spots (Pi, Qj) successfully matched by the Hungarian algorithm ij :

[0174] ΔI ij =Intensity(Q j )-Intensity(P i )

[0175] Among them, Intensity (Q j ) represents protein spot Q j Intensity (Intensity value / Grayscale value), Intensity (P i ) represents protein dot P i intensity (intensity value / grayscale value).

[0176] (22) Statistical analysis

[0177] Calculate the mean intensity change:

[0178]

[0179] Where K represents the number of protein point pairs that are successfully matched between two adjacent image frames determined by the Hungarian algorithm;

[0180] Intensity change distribution: Display the intensity changes of individual protein spots through histograms or box plots.

[0181] (23) Example

[0182] Among the 90 successfully matched protein spots, the average fluorescence intensity changes from 500 (Frame 1) to 520 (Frame 2):

[0183] Average strength increase

[0184] (3) Real-time calculation of the area change of protein spots (Area Change) is achieved by the following method:

[0185] (31) Area calculation

[0186] Since the number and position of pixels contained in each protein spot can be determined by image region segmentation, and the size and area of ​​the pixels are certain, the pixel area of ​​each protein spot can be calculated by binarization and connected domain analysis.

[0187] (32) Matching point pairs

[0188] Assume that the protein points that are successfully matched in two adjacent frames of images are (Pi, Qj) determined by the Hungarian algorithm;

[0189] Calculate the area change ΔA for the successfully matched protein spots ij :

[0190] ΔA ij =Area(Q j )-Area(P i )

[0191] Among them, Area(Q j ) represents protein spot Q j The pixel area, Area(P i ) represents protein dot P i pixel area.

[0192] (33) Statistical analysis

[0193] Calculate the average area change:

[0194]

[0195] Here, K represents the number of protein point pairs that are successfully matched in two adjacent image frames determined by the Hungarian algorithm.

[0196] Calculate the area expansion / contraction ratio:

[0197] Expansion Ratio=Area(Q j ) / Area(P i )

[0198] When the Expansion Ratio is greater than 1, it is the expansion ratio. When the Expansion Ratio is less than 1, it is the contraction ratio. When the Expansion Ratio is equal to 1, the area remains unchanged.

[0199] (34) Example

[0200] Among the 90 protein spots that were successfully matched, the average area ranged from 50px to 2 (Frame 1) becomes 55px 2 (Frame 2):

[0201] Average area increase

[0202] Expansion ratio = 55 / 50 = 1.1 (ie, the area increases by 10%).

[0203] (4) Real-time calculation of the spatial distribution change of protein spots is achieved by the following method:

[0204] (41) Center of mass displacement

[0205] Calculate the moving distance ΔD for the successfully matched protein spots ij :

[0206] ΔD ij =||Position(D j )-Position(D i )||

[0207] Among them, Position(D j ) indicates that the protein points D that are successfully matched in two adjacent frames are determined by the Hungarian algorithm. j The center of mass position, Position(D i ) indicates that the protein points D that are successfully matched in two adjacent frames are determined by the Hungarian algorithm. i The center of mass position of ||Position(Qj)-Position(Pi)|| represents the protein point D that is successfully matched j Protein D i The distance between them.

[0208] (42) Overall distribution changes

[0209] Calculate the mean displacement:

[0210]

[0211] Here, K represents the number of protein point pairs that are successfully matched in two adjacent image frames determined by the Hungarian algorithm.

[0212] Directionality analysis: Calculate the directional consistency of the displacement vector (such as whether it migrates in a certain direction) to determine the migration direction of the protein spot.

[0213] (43) Density change

[0214] The spatial distribution differences of protein spots in the two frames were compared by kernel density estimation (KDE).

[0215] (44)Example

[0216] Among the 90 successfully matched protein spots, the average displacement distance was 5px, indicating that the protein spots as a whole moved slightly.

[0217] If the direction of displacement is concentrated in a certain area (e.g., to the left), it may indicate directional migration.

[0218] In some embodiments, in step S22, the shape factor of the detected protein spots is calculated using the following formula:

[0219]

[0220] Among them, Circularity represents the circularity, Area represents the area of ​​the protein dot, and Perimeter represents the perimeter of the protein dot.

[0221] Since the number and position of pixels contained in each protein dot can be determined by segmenting the image area, and the size and area of ​​the pixels are determined, the perimeter and area of ​​the protein dot can be calculated.

[0222] In some embodiments, in step S22 , the cluster analysis algorithm adopts the DBSCAN algorithm.

[0223] Several protein spots were classified using the DBSCAN algorithm.

[0224] For example, classification is performed based on the area of ​​protein spots. If the area of ​​a protein spot is within a certain preset area threshold range, it is considered to be the same protein; if it is within another preset area threshold range, it is considered to be another protein.

[0225] In some embodiments, in step S23, when visually displaying the analysis results, the analysis results are presented in the form of a chart using a preset visualization tool;

[0226] In step S23, the analysis report includes analysis data and charts, and supports export operations in several different formats.

[0227] In this embodiment, protein spot distribution heat maps, fluorescence intensity curves, shape parameter distribution maps, etc. can be generated through existing visualization tools.

[0228] In this embodiment, the analysis report can be exported to PDF or Excel format.

[0229] In some embodiments, the present invention can be applied to software interaction for protein spot identification and analysis, referring to Figure 6 , which is a flowchart of software interaction, specifically including the following steps:

[0230] (1) The user chooses to upload a picture.

[0231] (2) The front end loads the uploaded image.

[0232] (3) The chip template pops up.

[0233] When the hardware scans for protein spots, cells are placed into the evenly distributed wells of the experimental chip. Electrophoresis is used to rupture the cells, and the labeled proteins within the cells are displaced out of the chip's cavities. The chip template in this step refers to the distribution of the wells on the experimental chip.

[0234] (4) Select a template.

[0235] (5) Call the image enhancement algorithm.

[0236] Step S11 in the above embodiments of the present invention is used to pre-process the image.

[0237] Then, a wave of identification of the swimming lane hole position is performed; if the template selected in step (4) does not match the template in the image, the return value is 2; if the template selected in step (4) matches the template in the image but the image quality is relatively poor, that is, the position of the edge point cannot be determined, the return value is 0; otherwise, the preprocessing is considered successful and the return value is 1.

[0238] (6) When the return value obtained in step (5) is 0, the image enhanced by step (5) is first rendered, and then a dialog box pops up to allow the user to choose manual identification or to re-change the chip template parameters. If the user chooses to re-change the chip template parameters, the process returns to step (3). If the user chooses manual identification, a point selector pops up to select a point. The point selector can directly use the existing open source ImageJ toolkit, call the manual lane identification method, and then jump to step (8).

[0239] When the return value obtained in step (5) is 1, proceed to the next step.

[0240] When the return value obtained in step (5) is 2, it indicates that the image is illegal.

[0241] (7) The front-end renders the enhanced image and calls the algorithm for automatic lane identification.

[0242] (8) Render the lanes.

[0243] (9) Call the algorithm for automatic protein spot identification.

[0244] The picture is processed using steps S12 to S22 in the above embodiments of the present invention.

[0245] (10) Rendering protein lanes.

[0246] The lanes are rendered based on the positions of the protein spot centroids and lane wells identified by the automatic protein spot recognition algorithm.

[0247] The protein spots obtained by the above method can be used for various data analyses, such as Figure 4As shown in the figure, the black line is the source data, the horizontal axis represents the distance from the lane hole in a single lane containing a protein spot, and the vertical axis represents the average grayscale value within the lane width at that distance. The source data is Gaussian fitted to obtain the Gaussian fitting data, which is the green line, which is used to fit the irregular source data into a standard normal distribution. Figure 4 In the formula, Model refers to the fitting method, and Gaussian fitting is selected; Equation refers to the fitting equation, and existing technology can be used, for example, the formula: y=y0+(A / (w*sqrt(PI / 2)))*exp(-2*((x-xc) / w)^2); Plot refers to the number of peaks in the waveform graph; y0 represents the peak value after Gaussian fitting (the y-coordinate value corresponding to the highest point); xc represents the peak position after Gaussian fitting (the x-coordinate value corresponding to the highest point); w represents the half-peak width multiplied by the square root of 2ln2; A represents the peak area; Reduced Chi-Sqr represents the simplified chi-square; Square (CoD) represents R-square, also called the coefficient of determination; Adj.R-Square represents the adjusted R-square.

[0248] For example Figure 5 As shown, the horizontal axis represents the lane number, which is generated by the system and is numbered from left to right and from top to bottom. The vertical axis represents the center point, which is the value of the peak point after Gaussian fitting. Figure 4 The value of xc can also be understood as the position of the center of mass of the protein point. Figure 5 The scatter plots are basically between 300 and 400, indicating that the displacement distances of the observed proteins under the same intensity of electrophoresis are not much different. We can further analyze indicators such as peak value and / or area to determine the type of observed protein.

[0249] An embodiment of the present invention further provides a protein spot recognition and analysis device based on real-time cell imaging, the protein spot recognition and analysis device based on real-time cell imaging comprising:

[0250] The image preprocessing module is used to obtain cell fluorescence images, perform denoising on the cell fluorescence images using the non-local mean denoising algorithm, and use the morphological opening operation to remove background inhomogeneity to obtain the preprocessed image;

[0251] The protein spot detection module is used to dynamically calculate the threshold of the preprocessed image based on the local image intensity and perform binarization to obtain an image with background and foreground. The image is subjected to morphological operations to remove noise and artifacts, and the image is segmented to obtain an image with several protein spots.

[0252] Real-time data analysis module, used to match protein spots in multiple frames of images based on the Hungarian algorithm and calculate the dynamic parameters of protein spots in real time;

[0253] Multi-parameter statistical analysis module, used to calculate the shape factor of the detected protein spots, cluster the protein spots in the image using a clustering analysis algorithm, and label one or more types of protein spots;

[0254] The data visualization and report generation module is used to visualize the analysis results, including the dynamic parameters, shape factors, and types of protein spots, and generate analysis reports.

[0255] The present invention has been described in detail above with reference to the embodiments of the accompanying drawings. A person skilled in the art can make various modifications to the present invention based on the above description. Therefore, certain details in the embodiments should not be construed as limiting the present invention. The scope of protection of the present invention shall be determined by the scope defined by the appended claims.

Claims

1. A protein spot identification method based on real-time cell imaging, characterized in that: The protein spot identification method based on real-time cell imaging includes: S11, acquiring a cell fluorescence image, performing denoising on the cell fluorescence image using a non-local means denoising algorithm, and using a morphological opening operation to remove background inhomogeneity to obtain a preprocessed image; S12, dynamically calculating a threshold of the preprocessed image based on the local image intensity and binarizing it to obtain an image with a background and a foreground, removing noise and artifacts from the image through morphological operations, and segmenting the image to obtain an image with several protein spots.

2. The protein spot identification method based on real-time cell imaging according to claim 1, characterized in that: In step S11, the cell fluorescence image is denoised using a non-local means denoising algorithm, which is implemented using the following formula: Among them, x represents the position of the current pixel to be denoised, I denoised (x) represents the intensity value of pixel x in the denoised image, C(x) represents the normalization constant, y represents the position of another pixel in the image, Ω represents the set of all pixels y in the image that may participate in the calculation, w(x, y) is the weight function, which represents the contribution of pixel y to the denoising of pixel x, and I(y) represents the intensity value of pixel y in the original image; And / or, in step S11, a morphological opening operation is used to remove background inhomogeneity, which is implemented using the following formula: I corrected =I original –Opening(I original ) Among them, I corrected represents the image after background correction, I original Represents the original input image, Opening(I original ) represents the result of morphological opening operation on the original input image.

3. The protein spot identification method based on real-time cell imaging according to claim 1, characterized in that: In step S12, the pre-processed image is thresholded dynamically based on the local image intensity and binarized using the following formula: T(x)=μ(x)+k·σ(x) Where T(x) represents the dynamic threshold of pixel x, μ(x) represents the average grayscale value of the pixel x, σ(x) represents the standard deviation of the grayscale value of the pixel x, and k is the adjustment coefficient. According to the result of T(x), the pixels are divided into foreground and background to generate a binary image; And / or, in step S12, when removing noise and artifacts from the image, it is achieved through a morphological opening operation or a closing operation.

4. A protein spot analysis method based on real-time cell imaging, characterized in that: The protein spot analysis method based on real-time cell imaging includes: S21, matching the protein spots in the multiple frames of images obtained by the protein spot recognition method based on real-time cell imaging according to any one of claims 1 to 3 based on the Hungarian algorithm, and calculating the dynamic parameters of the protein spots in real time; S22, calculating the shape factor of the detected protein spots, clustering the protein spots in the image using a clustering analysis algorithm, and marking one or more types of protein spots; S23, the analysis results including the dynamic parameters, shape factors and types of the protein spots are visualized and an analysis report is generated.

5. The protein spot analysis method based on real-time cell imaging according to claim 4, characterized in that: In step S21, protein spots in multiple frames of images are matched based on the Hungarian algorithm, which is implemented using the following formula: Cost Matrix ij =||Position i -Position j ||+λ||Intensity i -Intensity j || Among them, Cost Matrix ij represents the matching cost between the i-th protein point and the j-th protein point, ||Position i -Position j || represents the position difference between the i-th protein spot and the j-th protein spot, λ represents the weight coefficient, and ||Intensity i -Intensity j || represents the intensity difference between the i-th protein spot and the j-th protein spot.

6. The protein spot analysis method based on real-time cell imaging according to claim 4, characterized in that: In step S21, the dynamic parameters of the protein spots include at least one of a change in the number of protein spots, a change in the fluorescence intensity of the protein spots, a change in the area of ​​the protein spots, and a change in the distribution of the protein spots; Real-time calculation of the number changes of protein spots is achieved by the following method: Assume that the protein point sets in two adjacent frames of images are {P1, P2, ..., PN} and {Q1, Q2, ..., QM} respectively; The Hungarian algorithm is used to determine the number of successfully matched protein spots in two adjacent image frames as K; The following quantity change calculation is performed on the protein spots in two adjacent frames of images: New protein point = MK Disappearing protein spots = NK ΔN=MN ΔN represents the net change of protein spots in two adjacent frames. When ΔN>0, it means the number of protein spots increases; when ΔN<0, it means the number of protein spots decreases; when ΔN=0, it means the number of protein spots remains unchanged. Real-time calculation of the fluorescence intensity changes of protein spots is achieved by the following method: Assume that the protein points that are successfully matched in two adjacent frames of images are (Pi, Qj) determined by the Hungarian algorithm; Calculate the intensity difference ΔI for the successfully matched protein spots ij : ΔI ij =Intensity(Q j )-Intensity(P i ) Among them, Intensity (Q j ) represents protein spot Q j Intensity (P i ) represents protein dot P i Strength; Calculate the mean intensity change: Where K represents the number of protein point pairs that are successfully matched between two adjacent image frames determined by the Hungarian algorithm; Real-time calculation of the area change of protein spots is achieved by the following method: Assume that the protein points that are successfully matched in two adjacent frames of images are (Pi, Qj) determined by the Hungarian algorithm; Calculate the pixel area of ​​each protein spot; Calculate the area change ΔA for the successfully matched protein spots ij : ΔA ij =Area(Q j )-Area(P i ) Among them, Area(Q j ) represents protein spot Q j The pixel area, Area(P i ) represents protein dot P i The pixel area; Calculate the average area change: Where K represents the number of protein point pairs that are successfully matched between two adjacent image frames determined by the Hungarian algorithm; Calculate the area expansion / contraction ratio: Expansion Ratio=Area(Q j ) / Area(P i ) Real-time calculation of protein spot distribution changes is achieved by: Calculate the moving distance ΔD for the successfully matched protein spots ij : ΔD ij =||Position(D j )-Position(D i )|| Among them, Position(D j ) indicates that the protein points D that are successfully matched in two adjacent frames are determined by the Hungarian algorithm. j The center of mass position, Position(D i ) indicates that the protein points D that are successfully matched in two adjacent frames are determined by the Hungarian algorithm. i The center of mass position of ||Position(Qj)-Position(Pi)|| represents the protein point D that is successfully matched j Protein D i the distance between them; Calculate the mean displacement: Here, K represents the number of protein point pairs that are successfully matched in two adjacent image frames determined by the Hungarian algorithm.

7. The protein spot analysis method based on real-time cell imaging according to claim 6, characterized in that: Calculate the intensity difference ΔI for the successfully matched protein spots ij Finally, the intensity changes of individual protein spots are displayed through histograms or box plots; And / or, calculate the moving distance ΔD of the successfully matched protein spots ij Finally, the migration direction of the protein spots is determined; And / or, after calculating the distribution changes of the protein spots in real time, the spatial distribution differences between the two frames are compared by kernel density estimation.

8. The protein spot analysis method based on real-time cell imaging according to claim 4, wherein In step S22, the shape factor of the detected protein spots is calculated using the following formula: Among them, Circularity represents the circularity, Area represents the area of ​​the protein dot, and Perimeter represents the perimeter of the protein dot; And / or, in step S22, the cluster analysis algorithm adopts the DBSCAN algorithm.

9. The protein spot analysis method based on real-time cell imaging according to claim 4, wherein: In step S23, when visualizing the analysis results, the analysis results are presented in the form of a chart using a preset visualization tool; In step S23, the analysis report includes analysis data and charts, and supports export operations in several different formats.

10. A protein spot recognition and analysis device based on real-time cell imaging, characterized in that: The protein spot recognition and analysis device based on real-time cell imaging comprises: An image preprocessing module is used to obtain a cell fluorescence image, perform denoising on the cell fluorescence image using a non-local means denoising algorithm, and use a morphological opening operation to remove background inhomogeneity to obtain a preprocessed image; The protein spot detection module is used to dynamically calculate the threshold of the preprocessed image based on the local image intensity and perform binarization to obtain an image with background and foreground. The image is then subjected to morphological operations to remove noise and artifacts and image region segmentation to obtain an image with several protein spots. Real-time data analysis module, used to match protein spots in multiple frames of images based on the Hungarian algorithm and calculate the dynamic parameters of protein spots in real time; Multi-parameter statistical analysis module, used to calculate the shape factor of the detected protein spots, cluster the protein spots in the image using a clustering analysis algorithm, and label one or more types of protein spots; The data visualization and report generation module is used to visualize the analysis results, including the dynamic parameters, shape factors, and types of protein spots, and generate analysis reports.

Images