Vesicle size analysis method and system based on artificial intelligence

Abstract

The invention discloses a vesicle size analysis method and system based on artificial intelligence, and relates to the technical field of biomedical research, and the method comprises the steps: constructing a preliminary graph structure through a Sobel operator, enabling the preliminary graph structure to be improved in edge extraction precision, effectively processing the space interaction and dense distribution between vesicles, and improving the edge extraction efficiency. By means of semantic segmentation of the U-Net deep neural network, the automation degree and consistency of vesicle recognition are improved, manual intervention and subjective errors are reduced, fine division of vesicle overlapping or dense region boundaries is achieved by inputting a mask graph generated by the U-Net and a constructed graph structure into the graph neural network, and the accuracy of vesicle recognition is improved. And the accuracy and repeatability of vesicle size calculation are ensured.

Description

Technical Field

[0001] This invention relates to the field of biomedical research technology, and in particular to an artificial intelligence-based method and system for vesicle size analysis. Background Technology

[0002] In recent years, with the widespread application of extracellular vesicles in early disease diagnosis, biomarker research, and drug delivery, vesicle size analysis has become crucial for understanding intracellular and extracellular substance transport mechanisms and the development of disease diagnosis and drug delivery systems. Transmission electron microscopy (TEM) is widely used for vesicle structure observation due to its high resolution and rich image detail. However, traditional vesicle size measurement methods typically rely on manual image interpretation, which suffers from high subjectivity, low efficiency, and poor reproducibility, making it difficult to meet the needs of large-scale data analysis and high-throughput clinical applications. However, with a deeper understanding of vesicle function and its heterogeneity, researchers increasingly need a method that can efficiently and accurately process large amounts of data. In recent years, with the rapid development of artificial intelligence technologies such as computer vision, image processing, and deep learning, researchers have begun to attempt to use automated algorithms to identify and analyze vesicle images, such as using edge detection, image segmentation, and clustering classification to extract and measure vesicle size. However, existing methods mostly focus on the two-dimensional image level and fail to fully combine the representation capabilities of graph structures and neural networks, resulting in insufficient recognition accuracy in densely distributed vesicle areas and poor handling of blurred edge areas, which limits their application value in high-throughput and high-precision measurement scenarios.

[0003] Existing vesicle image analysis methods face significant challenges when dealing with complex backgrounds, overlapping regions, and scenes with blurred boundaries. Firstly, traditional image processing algorithms (such as edge detection and thresholding) often struggle to accurately extract vesicle edges in images with diverse vesicle morphologies and significant imaging noise, leading to oversegmentation or undersegmentation. Secondly, while deep learning-based image segmentation methods demonstrate strong feature extraction capabilities in medical images, they still suffer from insufficient accuracy in edge recognition within high-density vesicle regions. Furthermore, existing methods often neglect the spatial correlation and structural relationships of vesicles within images, failing to effectively address issues such as blurred boundary fusion or size misestimation between vesicles. Therefore, there is an urgent need to achieve fine segmentation and accurate size quantization of vesicle boundaries in complex images. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an artificial intelligence-based vesicle size analysis method and system to solve the problems of fine segmentation and accurate size quantification of vesicle boundaries in complex images.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an artificial intelligence-based vesicle size analysis method, which includes,

[0008] Acquire vesicle images and extract vesicle edge contours from the images using the Sobel operator to construct a graph structure;

[0009] Using vesicle images, the U-Net deep neural network is used to segment the vesicle region. The graph structure is combined with the graph neural network to optimize the boundary judgment of dense regions of vesicles and to calculate the vesicle size.

[0010] The K-means clustering algorithm was used to classify vesicle size data, generate distribution maps and heatmaps for visualization, and then organize and archive the data.

[0011] As a preferred embodiment of the artificial intelligence-based vesicle size analysis method of the present invention, the vesicle image acquisition refers to obtaining a fresh biological liquid sample, performing preliminary processing to remove cell debris and large particulate impurities, retaining the supernatant, transferring it to an ultracentrifuge for centrifugation, removing the supernatant after centrifugation, retaining the precipitate, resuspending the precipitate with sterile PBS buffer to obtain a clear vesicle suspension, adding it dropwise to the surface of a carbon membrane copper mesh, allowing it to stand to allow the vesicles to adsorb, absorbing excess liquid with filter paper, and allowing it to dry naturally at room temperature. Phosphotungstic acid solution is added to cover the carbon membrane copper mesh for negative staining, and the dried copper mesh is placed in a transmission electron microscope with standard imaging parameters to acquire vesicle images.

[0012] As a preferred embodiment of the artificial intelligence-based vesicle size analysis method of the present invention, the step of extracting vesicle edge contours from images using the Sobel operator to construct a graph structure involves preprocessing and enhancing vesicle images acquired by transmission electron microscopy, applying the Sobel edge detection operator to the enhanced images, extracting gradient maps in the horizontal and vertical directions respectively, calculating the gradient magnitude to obtain edge intensity images, extracting closed contour regions based on the edge intensity images, defining the geometric center point of each contour as a graph node, and calculating the Euclidean distance d between all nodes. ij ;

[0013] Set threshold When pixel distance d ij Greater than or equal to the threshold If the condition is met, then the relationship is not considered proximity and does not constitute an edge connection structure of the graph; otherwise, the relationship is considered proximity and constitutes an edge connection structure of the graph, thus constructing a node adjacency graph.

[0014] As a preferred embodiment of the artificial intelligence-based vesicle size analysis method of the present invention, the step of segmenting the vesicle region using a U-Net deep neural network through the vesicle image refers to uniformly cropping the processed grayscale image, performing normalization processing, and inputting it into the U-Net model for semantic segmentation to obtain a binary mask image.

[0015] As a preferred embodiment of the artificial intelligence-based vesicle size analysis method of the present invention, wherein: the step of combining graph structure to optimize the dense region vesicle boundary judgment using graph neural network and calculating vesicle size refers to connecting the binary mask image with the adjacent nodes. Figure 1 Using the same input graph neural network, feature vectors of graph nodes are constructed by utilizing the feature propagation mechanism between nodes. The optimized segmentation mask is obtained by updating node features using a graph neural network.

[0016] The optimized segmentation mask image is overlaid and visualized with the original transmission electron microscope image. The number of pixels occupied by each vesicle region in the image is used as the image area of ​​the vesicle. Combined with the image scale, this area is converted to an area A′ in actual units. n The equivalent diameter D of the vesicle is calculated by reverse calculation using the formula for the area of ​​a circle. n ;

[0017] Calculate the equivalent diameter of all vesicles to construct a vesicle size dataset D, and record the analysis results for each vesicle.

[0018] As a preferred embodiment of the artificial intelligence-based vesicle size analysis method of the present invention, the step of classifying vesicle size data using the K-means clustering algorithm refers to automatically classifying the vesicle size dataset D using the K-means clustering algorithm. The goal of the clustering process is to minimize the within-class squared error (WCSS) so that vesicles with similar sizes in the same class are assigned to the same category.

[0019] As a preferred embodiment of the artificial intelligence-based vesicle size analysis method of the present invention, the following steps are included: generating distribution maps and heatmaps for visualization and archiving: displaying the frequency distribution of the equivalent diameter of vesicles through a diameter histogram, sorting the equivalent diameter of each vesicle in descending order to obtain the median, maximum, and minimum values ​​of the equivalent diameter of vesicles; visually reflecting the dispersion of the population by displaying the median, maximum, and minimum values ​​of the vesicle diameter; superimposing the vesicle number, corresponding diameter, and cluster label back into the original image through a numbered labeling map; color-coding the image based on the size category of vesicles in the image coordinates through a spatial distribution heatmap to display the spatial arrangement patterns of different vesicle types; and uniformly organizing and saving the statistical results.

[0020] Secondly, the present invention provides an artificial intelligence-based vesicle size analysis system, comprising,

[0021] The image acquisition module is used to extract vesicles from biological fluids by centrifugation, and then acquire images using a transmission electron microscope after negative staining.

[0022] The graph structure module is used for image grayscale conversion, noise reduction and enhancement, edge extraction and construction of node adjacency graph structure;

[0023] The mask image generation module is used to segment images using the U-Net model and output vesicle binary mask images;

[0024] The graph optimization and size calculation module is used to combine graph neural networks to optimize boundary recognition and calculate the equivalent diameter data of vesicles.

[0025] The clustering analysis and archiving module is used to cluster and classify size data, generate charts, and organize and archive statistical results.

[0026] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the artificial intelligence-based vesicle size analysis method described in the first aspect of the present invention.

[0027] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the artificial intelligence-based vesicle size analysis method as described in the first aspect of the present invention.

[0028] The beneficial effects of this invention are as follows: This invention constructs a preliminary graph structure through the Sobel operator, thereby improving the accuracy of edge extraction and effectively handling the spatial interaction and dense distribution between vesicles. Through semantic segmentation using the U-Net deep neural network, the automation and consistency of vesicle recognition are improved, reducing human intervention and subjective errors. By inputting the mask image generated by U-Net and the constructed graph structure into the graph neural network, fine division of the boundaries of overlapping or dense regions of vesicles is achieved, ensuring the accuracy and repeatability of vesicle size calculation. Attached Figure Description

[0029] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1This is a flowchart of an artificial intelligence-based vesicle size analysis method in Example 1.

[0031] Figure 2 This is a structural diagram of an artificial intelligence-based vesicle size analysis system in Example 1. Detailed Implementation

[0032] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0033] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0034] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0035] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides an artificial intelligence-based vesicle size analysis method, including the following steps:

[0036] S1. Acquire vesicle images and extract vesicle edge contours from the images using the Sobel operator to construct a graph structure;

[0037] Specifically, vesicle image acquisition involves obtaining fresh biological liquid samples (cell culture supernatant, plasma, or urine), performing preliminary processing to remove cell debris and large particulate impurities, retaining the supernatant, and transferring it to an ultracentrifuge for centrifugation (using the centrifugal force experienced by an object in a rotating system to separate the mixture). After centrifugation, the supernatant is removed, and the precipitate (i.e., the crude vesicle extract) is retained. The precipitate is resuspended using sterile PBS buffer (filtered and sterilized phosphate buffer), and the vesicles are gently agitated to avoid rupture, resulting in a clear vesicle suspension. This suspension is then dropped onto the surface of a carbon membrane copper mesh, allowed to stand to allow vesicle adsorption, and excess liquid is absorbed using filter paper. The mesh is then allowed to dry naturally at room temperature. Phosphotungstic acid solution (a negative staining agent) is added to cover the carbon membrane copper mesh for negative staining, enhancing the electron density difference between the vesicle edge and the background and improving imaging contrast. The dried copper mesh is then placed in a transmission electron microscope (TEM) with standard imaging parameters (accelerating voltage, resolution, magnification, image saving format, and the inclusion of a scale bar in the image) to acquire vesicle images.

[0038] By acquiring fresh biological liquid samples and performing preliminary processing, the impact of impurities on the analytical results was reduced. Ultracentrifugation improved the purity of vesicle extraction. Resuspending the precipitate and adsorbing it onto a carbon membrane copper grid ensured the integrity and uniform distribution of the vesicles, improving sample quality in subsequent TEM imaging and reducing vesicle loss and positional shift. Negative staining and imaging preparation improved the quality of the imaging images, making the edges of the vesicles more distinct and ensuring resolution and accuracy in subsequent analysis.

[0039] Furthermore, by extracting vesicle edge contours from the images using the Sobel operator to construct a graph structure, the vesicle images acquired by the transmission electron microscope are uniformly converted into grayscale images, and Gaussian filtering is applied for noise reduction and CLAHE local contrast enhancement. To obtain the preliminary boundary contours of the vesicles, the Sobel edge detection operator is applied to the enhanced images to extract gradient maps in the horizontal and vertical directions, and the gradient magnitudes are calculated to obtain the edge intensity images.

[0040]

[0041] Where G is the final edge intensity image, representing the edge sharpness. x G y These are gradient maps in the horizontal and vertical directions, respectively.

[0042] Closed contour regions are extracted based on edge intensity images, and the geometric center point of each contour is defined as a graph node. The Euclidean distance between all nodes is then calculated.

[0043]

[0044] Where, d ij It is the pixel distance between node i and node j, (x i ,y i ), (x j ,y j () represents the image coordinates of the centers of the two vesicles;

[0045] Thresholds were set using cluster analysis. When pixel distance d ij Greater than or equal to the threshold If the condition is met, then the relationship is not considered proximity and does not constitute an edge connection structure of the graph; otherwise, the relationship is considered proximity and constitutes an edge connection structure of the graph, thus constructing a node adjacency graph.

[0046] By reducing noise and improving local contrast, image quality is ensured, making vesicle edges easier to detect and segment, thus improving overall analysis accuracy. Edge detection and gradient calculation using the Sobel operator effectively enhance the clarity of vesicle edges, making subsequent boundary extraction, segmentation, and size analysis more accurate and reliable. This provides high-quality data support for size measurement and morphological analysis. Closed contour extraction and graph node construction from edge intensity images provide an accurate foundation for subsequent graph analysis, clustering, and graph neural network optimization, ensuring data standardization and scientific analysis.

[0047] S2. Using vesicle images, the U-Net deep neural network is used to segment the vesicle region, and graph neural network optimization is performed to determine the boundary of dense vesicles and calculate the vesicle size.

[0048] Specifically, using vesicle images, the U-Net deep neural network is used to segment the vesicle regions. This involves uniformly cropping and normalizing the processed grayscale images to match the input requirements of the U-Net model. A training dataset is constructed using historically labeled transmission electron microscope images and input into the U-Net model. Data augmentation is performed through rotation, flipping, and scaling to improve the model's generalization ability. The encoder (downsampling) in the model structure extracts image features (edges, textures, structures) and reduces the resolution. The decoder (upsampling) restores the image size and fuses spatial information, outputting a binary mask of the vesicle regions. During training, a loss function is used to measure the similarity between the predicted image and the ground truth labeled image (labeled by experts). The Adam optimizer is used to update parameters, and the learning rate is dynamically adjusted using a learning rate decay strategy. Training continues until the maximum number of iterations is reached. The iteration stops, and the model parameters are output to update the U-Net model. The grayscale image is then input into the updated U-Net model for semantic segmentation to obtain a binary mask image (regions with a value of 1 represent identified vesicles, and regions with a value of 0 represent the background).

[0049] Image cropping and normalization ensured the quality of the input images, enabling the model to learn efficiently and accurately, improving subsequent segmentation accuracy, and reducing error sources in the data processing process. Building a training dataset and performing data augmentation increased the diversity of training data and improved adaptability to different situations (such as changes in location and scale). The encoder and decoder in the U-Net model structure improved the model's accuracy in vesicle region segmentation. A precise loss function and efficient optimization algorithm ensured the stability and efficiency of the model training process, achieving high-precision segmentation results. Using the updated model for semantic segmentation enabled accurate and automated segmentation of vesicle regions, greatly improving analysis efficiency and reducing human error.

[0050] Furthermore, by combining graph structure with graph neural network optimization for dense region vesicle boundary judgment and vesicle size calculation, the binary mask image is adjacent to the nodes. Figure 1 Using the same input graph neural network, the feature propagation mechanism between nodes is utilized to improve the accuracy of boundary recognition in dense regions and construct feature vectors for graph nodes. Among them, g i It is the mean gray-level gradient at the edge of the vesicle, c i This is the local contrast enhanced by CLAHE, (x i ,y i ) represents the spatial coordinates of the node;

[0051] An optimized segmentation mask is obtained by updating node features using a graph neural network, which achieves higher accuracy in areas with blurred boundaries or overlapping vesicles.

[0052]

[0053] in, It is the feature representation of node i at layer t+1. Let w be the feature representation of node j at layer t, N(i) be the set of neighboring nodes of node i, and w be the feature representation of node j at layer t. ij σ is the edge weight, which is set through experiments; W is the learnable weight matrix; and σ is the activation function.

[0054] The optimized segmentation mask is overlaid and visualized on the original transmission electron microscope image. The number of pixels occupied by each vesicle region in the optimized segmentation mask is used as the image area of ​​the vesicle. Combined with the image scale, this area is converted to its actual unit area using the following formula:

[0055] A′ n =A n ×s 2

[0056] Among them, A′ n A is the actual projected area of ​​the nth vesicle. n 's' is the pixel area, and 's' is the scale bar.

[0057] The equivalent diameter of the vesicle is calculated by reverse calculation using the formula for the area of ​​a circle:

[0058]

[0059] Among them, D n A′ is the equivalent diameter of the nth vesicle. n It is the actual projected area of ​​the nth vesicle;

[0060] Calculate the equivalent diameter of all vesicles to construct a vesicle size dataset D = {D1, D2, ..., D...}n The analysis results (vesicle number, image coordinates, and equivalent diameter) for each vesicle are recorded.

[0061] By aligning the binary mask image with the node Figure 1 By using the same input graph neural network and leveraging the feature propagation mechanism between nodes, the segmentation accuracy of vesicles in complex environments is effectively improved, especially in cases of unclear boundaries or overlapping vesicles, ensuring high-quality segmentation results. By constructing feature vectors for graph nodes, the graph neural network can more efficiently capture the boundary information of vesicles. Through node feature updates and segmentation mask optimization, segmentation accuracy can be improved in complex backgrounds and overlapping regions. By calculating the actual projected area and equivalent diameter of vesicles, the size data of each vesicle can be provided, thereby constructing a vesicle size dataset. By constructing the vesicle size dataset and recording the analysis results, reliable data support can be provided for future research and applications.

[0062] S3. Use the K-means clustering algorithm to classify vesicle size data, generate distribution maps and heatmaps for visualization, and then organize and archive the data.

[0063] Specifically, classifying vesicle size data using the K-means clustering algorithm refers to using the K-means clustering algorithm on the vesicle size dataset D = {D1, D2, ..., D...}. n Automatic classification is performed to discover representative vesicle size subpopulations. The goal of the clustering process is to minimize the within-cluster squared error (WCSS).

[0064]

[0065] Where k is the number of clusters (automatically determined according to the "elbow method"), and C q It is the q-th cluster category, μ q D is the average diameter of the q-th cluster category. n It is the equivalent diameter of vesicles belonging to this category;

[0066] By minimizing WCSS, vesicle sizes within the same class are made similar, enhancing the rationality of classification and assigning them to a single category.

[0067] By employing the K-means clustering algorithm to automatically classify vesicle size datasets, systematic data support is provided for further biological analysis and clinical research. The clustering process is optimized by minimizing the within-cluster squared error (WCSS), ensuring the consistency of vesicle size within clusters and enhancing classification accuracy.

[0068] Furthermore, the generation and visualization of distribution maps and heatmaps, along with their organization and archiving, involve using diameter histograms to display the frequency distribution of vesicle equivalent diameters, reflecting their concentration and skewness. The equivalent diameters of each vesicle are sorted in descending order to obtain the median, maximum, and minimum values. Displaying these values ​​visually reflects the dispersion of the population. A numbered labeling map overlays vesicle numbers, corresponding diameters, and cluster labels back onto the original image, facilitating manual verification of identification results and local structural analysis. A spatial distribution heatmap, based on the size category of vesicles in the image coordinates, color-codes the image to display the spatial arrangement patterns of different vesicle types. The statistical results are then uniformly organized and saved.

[0069] By displaying the frequency distribution of vesicle equivalent diameters using diameter histograms, researchers can clearly and intuitively understand the distribution characteristics of vesicle size. The ranking and statistical calculation of vesicle equivalent diameters make the dispersion and changing trends of the vesicle population more explicit. The use of numbered labeling maps for verification and identification allows researchers to easily validate and adjust analysis results, reducing human error and improving the reliability of the research process. The generation of spatial distribution heatmaps helps reveal functional differences of vesicles in different regions or their potential role in disease biomarker screening. The organization and storage of statistical results improves data management efficiency and enhances data traceability and reusability.

[0070] This embodiment also provides an artificial intelligence-based vesicle size analysis system, including:

[0071] The image acquisition module is used to extract vesicles from biological fluids by centrifugation, and then acquire images using a transmission electron microscope after negative staining.

[0072] The graph structure module is used for image grayscale conversion, noise reduction and enhancement, edge extraction and construction of node adjacency graph structure;

[0073] The mask image generation module is used to segment images using the U-Net model and output vesicle binary mask images;

[0074] The graph optimization and size calculation module is used to combine graph neural networks to optimize boundary recognition and calculate the equivalent diameter data of vesicles.

[0075] The clustering analysis and archiving module is used to cluster and classify size data, generate charts, and organize and archive statistical results.

[0076] This embodiment also provides a computer device applicable to an artificial intelligence-based vesicle size analysis method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the artificial intelligence-based vesicle size analysis method proposed in the above embodiment.

[0077] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0078] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements an artificial intelligence-based vesicle size analysis method as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

Claims

1. A method for vesicle size analysis based on artificial intelligence, characterized in that: include, Acquire vesicle images and extract vesicle edge contours from the images using the Sobel operator to construct a graph structure; Using vesicle images, the U-Net deep neural network is used to segment the vesicle region. The graph structure is combined with the graph neural network to optimize the boundary judgment of dense regions of vesicles and to calculate the vesicle size. The K-means clustering algorithm was used to classify vesicle size data, generate distribution maps and heatmaps for visualization, and then organize and archive the data.

2. The artificial intelligence-based vesicle size analysis method as described in claim 1, characterized in that: The acquisition of vesicle images refers to obtaining fresh biological liquid samples, performing preliminary processing to remove cell debris and large particulate impurities, retaining the supernatant, transferring them to an ultracentrifuge for centrifugation, removing the supernatant after centrifugation, retaining the precipitate, resuspending the precipitate with sterile PBS buffer to obtain a clear vesicle suspension, adding it dropwise to the surface of a carbon membrane copper mesh, allowing it to stand to allow vesicles to adsorb, absorbing excess liquid with filter paper, and allowing it to dry naturally at room temperature. Phosphotungstic acid solution is then added to cover the carbon membrane copper mesh for negative staining. The dried copper mesh is placed in a transmission electron microscope, and standard imaging parameters are configured to acquire vesicle images.

3. The artificial intelligence-based vesicle size analysis method as described in claim 2, characterized in that: The process of extracting vesicle edge contours from images using the Sobel operator to construct a graph structure involves preprocessing and enhancing vesicle images acquired by a transmission electron microscope, applying the Sobel edge detection operator to the enhanced images, extracting gradient maps in the horizontal and vertical directions, calculating the gradient magnitudes to obtain edge intensity images, extracting closed contour regions based on the edge intensity images, defining the geometric center point of each contour as a graph node, and calculating the Euclidean distance d between all nodes. ij ; Set threshold When pixel distance d ij Greater than or equal to the threshold If the condition is met, then the relationship is not considered proximity and does not constitute an edge connection structure of the graph; otherwise, the relationship is considered proximity and constitutes an edge connection structure of the graph, thus constructing a node adjacency graph.

4. The artificial intelligence-based vesicle size analysis method as described in claim 3, characterized in that: The process of segmenting the vesicle region using the U-Net deep neural network through the vesicle image involves uniformly cropping the processed grayscale image, normalizing it, and inputting it into the U-Net model for semantic segmentation to obtain a binary mask image.

5. The artificial intelligence-based vesicle size analysis method as described in claim 4, characterized in that: The method of combining graph structure with graph neural network optimization for dense region vesicle boundary judgment and vesicle size calculation involves inputting a binary mask image and a node adjacency graph into the graph neural network, and utilizing the feature propagation mechanism between nodes to construct feature vectors for graph nodes. The optimized segmentation mask is obtained by updating node features using a graph neural network. The optimized segmentation mask image is overlaid and visualized with the original transmission electron microscope image. The number of pixels occupied by each vesicle region in the image is used as the image area of ​​the vesicle. Combined with the image scale, this area is converted to an area A′ in actual units. n The equivalent diameter D of the vesicle is calculated by reverse calculation using the formula for the area of ​​a circle. n ; Calculate the equivalent diameter of all vesicles to construct a vesicle size dataset D, and record the analysis results for each vesicle.

6. The artificial intelligence-based vesicle size analysis method as described in claim 5, characterized in that: The classification of vesicle size data using the K-means clustering algorithm refers to the automatic classification of the vesicle size dataset D using the K-means clustering algorithm. The goal of the clustering process is to minimize the within-cluster squared error (WCSS) so that vesicles with similar sizes in the same cluster are assigned to the same category.

7. The artificial intelligence-based vesicle size analysis method as described in claim 6, characterized in that: The generation of distribution maps and heatmaps, visualization, organization, and archiving refers to displaying the frequency distribution of vesicle equivalent diameters using a diameter histogram, sorting the equivalent diameters of each vesicle in descending order, obtaining the median, maximum, and minimum values ​​of the vesicle equivalent diameters, and visually reflecting the dispersion of the population by displaying the median, maximum, and minimum values ​​of the vesicle diameters. A numbered labeling map is used to overlay the vesicle number, corresponding diameter, and cluster label back onto the original image. A spatial distribution heatmap is used, based on the size category of the vesicles in the image coordinates, to color-code the image, displaying the spatial arrangement patterns of different vesicle types. The statistical results are then uniformly organized and saved.

8. An artificial intelligence-based vesicle size analysis system, based on the artificial intelligence-based vesicle size analysis method according to any one of claims 1 to 7, characterized in that: include, The image acquisition module is used to extract vesicles from biological fluids by centrifugation, and then acquire images using a transmission electron microscope after negative staining. The graph structure module is used for image grayscale conversion, noise reduction and enhancement, edge extraction and construction of node adjacency graph structure; The mask image generation module is used to segment images using the U-Net model and output vesicle binary mask images; The graph optimization and size calculation module is used to combine graph neural networks to optimize boundary recognition and calculate the equivalent diameter data of vesicles. The clustering analysis and archiving module is used to cluster and classify size data, generate charts, and organize and archive statistical results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the artificial intelligence-based vesicle size analysis method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the artificial intelligence-based vesicle size analysis method according to any one of claims 1 to 7.

Images