A method and system for measuring intima-media thickness of blood vessels based on neural networks and shortest path method

Abstract

This invention provides a method and system for measuring vascular intima-media thickness based on neural networks and the shortest path algorithm. The method includes: acquiring and preprocessing an ultrasound image of the vessel to be measured; inputting the preprocessed ultrasound image into a trained neural network model to obtain the boundary probability map and orientation field map output by the neural network model; constructing a graph structure using pixels of the ultrasound image as nodes and adjacency relationships between pixels as edges, and dynamically assigning weights to the edges in the graph based on the boundary probability map and orientation field map to obtain a weighted graph; setting a start point and an end point for the vessel boundary to be searched in the weighted graph, and searching for the vessel boundary path based on the shortest path algorithm; calculating the average distance between the lumen-intima boundary and the media-adventitia boundary based on the vessel boundary path, and outputting the vascular intima-media thickness measurement result; which is beneficial for improving accuracy and robustness, has a high degree of automation, and good repeatability.

Description

Technical Field

[0001] This invention relates to the field of ultrasound medical technology, and in particular to a method and system for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method. Background Technology

[0002] Currently, atherosclerosis and its associated cardiovascular diseases (ASCVD) remain the leading cause of death and disability worldwide and in my country. Targeted early screening and prevention are essential throughout the entire lifespan. Emphasizing early screening and standardized treatment can effectively reduce the incidence and mortality of ASCVD. Intima-media thickness (IMT) is an important tool for assessing the risk of atherosclerosis and cardiovascular disease. Early detection of subclinical atherosclerosis can assess the risk of future cardiovascular and cerebrovascular events and monitor disease progression and treatment effectiveness. Currently, IMT measurement mainly relies on medical imaging equipment such as ultrasound and MRI. Ultrasound imaging is a non-invasive imaging technology widely used in medical diagnosis and monitoring. Due to its unique safety, real-time performance, and versatility, ultrasound examination has become an important diagnostic tool in modern medicine.

[0003] However, existing automatic or semi-automatic IMT measurement methods have the following limitations: First, traditional image processing methods (such as edge detection) rely on manually designed gradients, textures, and other features, which can easily lead to boundary loss or misjudgment when the image quality is poor, there is noise, or there are plaque lesions, resulting in poor robustness. The results also often depend on the selection of the initial contour. Second, pure deep learning segmentation methods (such as U-Net): While semantic segmentation models based on U-Net or its variants can output segmentation results end-to-end, their output is a pixel-level probability map, and the final boundary formation depends on post-processing (such as thresholding and contour search). This can lead to discontinuous boundaries, poor smoothness, and in some cases, unreasonable segmentation results. Finally, traditional graph search methods (such as the shortest path method): require manually defining the cost function of graph nodes (such as based on gray-level gradients), which is insufficiently adaptable to complex lesions and has weak generalization ability.

[0004] Therefore, in order to overcome the above-mentioned technical problems, the present invention provides a method and system for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method. Summary of the Invention

[0005] This invention provides a method and system for measuring intima-media thickness (IMMT) of blood vessels based on neural networks and the shortest path method. The method involves acquiring and preprocessing an ultrasound image of the blood vessel to be measured; inputting the preprocessed ultrasound image into a trained neural network model to obtain the boundary probability map and orientation field map output by the model; constructing a graph structure using pixels of the ultrasound image as nodes and adjacency relationships between pixels as edges; dynamically assigning weights to the edges in the graph based on the boundary probability map and orientation field map to obtain a weighted graph; setting a start and end point for the blood vessel boundary to be searched in the weighted graph; and then, based on the shortest path method... The short path algorithm searches for the vascular boundary path; based on the vascular boundary path, the average distance between the lumen-intima boundary and the media-adventitia boundary is calculated, and the intima-media thickness measurement result is output. This improves accuracy and robustness, overcomes the limitations of manually designed cost functions, and performs more stably and accurately under complex conditions such as low image quality, plaques, and noise interference. It also improves performance, has a high degree of automation and good repeatability, and has the potential to be applied to portable devices. Through lightweight model design and optimization, it can achieve rapid bedside or community screening in handheld ultrasound devices with limited computing power.

[0006] A method for measuring vascular intima-media thickness based on neural networks and the shortest path method includes: Acquire ultrasound images of the blood vessels to be measured and perform preprocessing; The preprocessed vascular ultrasound image is input into the trained neural network model to obtain the boundary probability map and orientation field map output by the neural network model. A graph structure is constructed using pixels of vascular ultrasound images as nodes and adjacency relationships between pixels as edges. Weights are dynamically assigned to the edges in the graph based on boundary probability maps and orientation field maps to obtain a weighted graph. In the weighted graph, a start point and an end point are set for the blood vessel boundary to be searched, and the blood vessel boundary path is obtained based on the shortest path algorithm; Based on the vascular boundary path, the average distance between the lumen-intima boundary and the media-adventitia boundary is calculated, and the intima-media thickness measurement results are output.

[0007] Preferably, a method for measuring intima-media thickness based on neural networks and the shortest path method, wherein the training process of the neural network model includes: Obtain a training dataset containing vascular ultrasound images and corresponding manually annotated lumen-intima and media-adventitia boundaries; After preprocessing the images in the training dataset, they are input into the initial neural network model for training. By optimizing the network parameters, the boundary probability map output by the model is made to approximate the true annotation, thus obtaining the trained neural network model.

[0008] Preferably, a method for measuring intimal media thickness based on neural networks and the shortest path method includes: In the step of dynamically assigning weights to edges in the graph based on the boundary probability map and the orientation field map, the weight W of the edge between the current pixel and its neighboring pixels is calculated using the following formula: ; in, This represents the boundary probability value of the corresponding pixel in the boundary probability map; This represents the difference between the direction value of the current pixel in the direction field map and the direction vector pointing to neighboring pixels; This represents the auxiliary cost term calculated based on the image grayscale gradient. , , These are the weighting coefficients.

[0009] Preferably, a method for measuring intimal media thickness based on neural networks and the shortest path method includes: Calculate the Using the following formula: ; in, This represents the value of the current pixel in the orientation field map; It represents the direction vector from the current center pixel to one of its neighboring pixels.

[0010] Preferably, a method for measuring intimal media thickness based on neural networks and the shortest path method includes: Calculate the Using the following formula: ; in, This represents the gradient magnitude of neighboring pixels calculated using the image gradient operator on the original grayscale image.

[0011] Preferably, a method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method, wherein the shortest path algorithm searches for the vascular boundary path, including: The Dijkstra algorithm is used to search for the minimum cost paths corresponding to the lumen-inner membrane boundary and the middle membrane-outer membrane boundary in the weighted graph.

[0012] Preferably, a method for measuring intima-media thickness based on neural networks and the shortest path method sets a start and end point for the vessel boundary to be searched in a weighted graph, including: Based on the lumen-intima boundary and the media-outer membrane boundary, in the boundary probability diagram, the node with the highest corresponding boundary probability value in the leftmost column of the ultrasound image is selected as the starting point of the corresponding weighted diagram, and the node with the highest corresponding boundary probability value in the rightmost column of the ultrasound image is selected as the ending point of the weighted diagram.

[0013] Preferably, a method for measuring vascular intima-media thickness based on neural networks and the shortest path method calculates the average distance between the lumen-intima boundary and the media-adventitia boundary, including: Obtain each pixel node on the path between the middle and outer membranes, and calculate the shortest vertical distance from each pixel to the path between the lumen and the inner membrane. Calculate the average of the shortest vertical distances of all pixel nodes as the average distance between the lumen-inner membrane boundary and the middle and outer membrane boundary.

[0014] Preferably, a method for measuring intima-media thickness based on neural networks and the shortest path method, using pixels of vascular ultrasound images as nodes and adjacency relationships between pixels as edges, includes the following: Each pixel node is connected to the pixel nodes in its eight-neighborhood to form an edge; wherein the eight-neighborhood includes adjacent pixel nodes in the top, bottom, left, right and four diagonal directions.

[0015] A system for measuring vascular intima-media thickness based on neural networks and the shortest path method, comprising: The image acquisition and preprocessing module is used to acquire and preprocess the ultrasound images of the blood vessels to be measured. The neural network feature extraction module is used to input the preprocessed vascular ultrasound image into the trained neural network model to obtain the boundary probability map and orientation field map output by the neural network model. The weighted graph construction module is used to construct a graph structure with pixels of vascular ultrasound images as nodes and the adjacency relationship between pixels as edges. It dynamically assigns weights to the edges in the graph based on the boundary probability map and the orientation field map to obtain a weighted graph. The boundary path search module is used to set the start and end points of the blood vessel boundary to be searched in the weighted graph, and to search for the blood vessel boundary path based on the shortest path algorithm. The thickness calculation and output module is used to calculate the average distance between the lumen-intima boundary and the media-adventitia boundary based on the vascular boundary path, and output the vascular intima-media thickness measurement results.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This method involves acquiring and preprocessing ultrasound images of the blood vessels to be measured; inputting the preprocessed ultrasound images into a trained neural network model to obtain the boundary probability map and orientation field map output by the neural network model; constructing a graph structure using pixels of the ultrasound images as nodes and adjacency relationships between pixels as edges, and dynamically assigning weights to the edges in the graph based on the boundary probability map and orientation field map to obtain a weighted graph; setting the start and end points for the blood vessel boundaries to be searched in the weighted graph, and searching for the blood vessel boundary paths based on the shortest path algorithm; calculating the average distance between the lumen-intima boundary and the media-adventitia boundary based on the blood vessel boundary paths, and outputting the intima-media thickness measurement results; improving accuracy and robustness, overcoming the limitations of manually designed cost functions, and exhibiting more stable and accurate performance under complex conditions such as low image quality, plaques, and noise interference; performance improvement; high degree of automation and good repeatability; potential for portable devices, enabling rapid bedside or community screening in handheld ultrasound devices with limited computing power through lightweight model design and optimization.

[0017] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method in an embodiment of the present invention; Figure 2 This is a neural network feature extraction diagram in a method for measuring vascular intima-media thickness based on neural networks and the shortest path method in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the construction of a weighted graph model in a method for measuring vascular intima-media thickness based on neural networks and the shortest path method, as described in an embodiment of the present invention. Figure 4 This is a flowchart of the boundary search based on the shortest path in a method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method in an embodiment of the present invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] Example 1: This example provides a method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method, such as... Figure 1 As shown, it includes: Step 1: Acquire and preprocess the ultrasound image of the blood vessel to be measured; Step 2: Input the preprocessed vascular ultrasound image into the trained neural network model to obtain the boundary probability map and orientation field map output by the neural network model; Step 3: Construct a graph structure using pixels of the vascular ultrasound image as nodes and adjacency relationships between pixels as edges, and dynamically assign weights to the edges in the graph based on the boundary probability map and orientation field map to obtain a weighted graph; Step 4: Set the start and end points for the blood vessel boundaries to be searched in the weighted graph, and search for the blood vessel boundary paths based on the shortest path algorithm; Step 5: Based on the vascular boundary path, calculate the average distance between the lumen-intima boundary and the media-adventitia boundary, and output the vascular intima-media thickness measurement results.

[0022] In this embodiment, the boundary probability map represents the probability that each pixel belongs to the lumen-intima boundary or the media-outer membrane boundary, and the orientation field map represents the orientation of the blood vessel wall in a local area of ​​the image.

[0023] In this embodiment, the blood vessel boundary path is a continuous path.

[0024] In this embodiment, preprocessing includes preprocessing the vascular ultrasound images in the dataset. During the data preprocessing stage, image standardization, noise reduction, and ROI (region of interest) cropping are performed to improve the efficiency and accuracy of model training.

[0025] In this embodiment, the preprocessed vascular ultrasound image is input into the trained network, which outputs two key pieces of information: a boundary probability map. The probability that each pixel belongs to the LII (luminal-inner membrane boundary) or MA (middle membrane-outer membrane boundary); orientation field map Predict the approximate orientation of the blood vessel walls in the image, specifically as follows: Figure 2 As shown.

[0026] In this embodiment, the image is treated as a graph structure, with the top-left corner of the image as the origin. Each pixel is a node, and the connection between a pixel and its neighboring pixels is an edge. The features output by the neural network are used to dynamically assign weights to each edge, specifically as follows: Figure 3 As shown.

[0027] In this embodiment, the above technical solution employs the following means: 1. Neural Network Feature Extraction: Using CNN (Convolutional Neural Network) to output boundary probability maps (the network learns and labels the probability that each pixel belongs to LII (luminal-intima boundary) and MA (media-adventitia boundary), with high-probability regions indicating that the boundary is more likely to be located there) and orientation field maps (the network learns and predicts the orientation of the blood vessel wall in local regions of the image, such as horizontal and vertical, providing a basis for the smoothness and continuity of the boundary), automatically identify blood vessel features, replace manual feature extraction, and enhance robustness in complex images.

[0028] 2. Dynamically Weighted Graph Construction: The weights of edges in the graph are dynamically calculated using the output of a neural network. The learning results of the neural network are transformed into constraints for graph search, building a bridge between the neural network and graph search to achieve intelligent search.

[0029] 3. Global Path Search: Apply Dijkstra's algorithm to search for the shortest path on the weighted graph, find the globally optimal, continuous, and smooth boundary, ensure the continuity of the results and the rationality of the dissection, and avoid fragmentation of the segmentation.

[0030] 4. IMT Calculation Output: Automated distance calculation enables end-to-end automatic measurement, outputting clinically usable results and improving the practicality and efficiency of the method.

[0031] The beneficial effects of the above technical solution are: higher accuracy and robustness, the neural network can adaptively learn complex image features and contextual information, automatically identify vascular features, overcome the limitations of manually designed cost functions, and perform more stably and accurately under complex conditions such as low image quality, plaques, and noise interference; improved performance, leveraging the advantages of neural networks in feature learning and the advantages of shortest path algorithms in structural optimization; high degree of automation, enabling fully automatic and high-precision IMT measurement without much manual intervention, with good repeatability; potential for portable devices, through lightweight model design and optimization, it can be deployed in handheld ultrasound devices with limited computing power to achieve rapid screening at the bedside or in the community.

[0032] Example 2: Based on Example 1, this example provides a method for measuring intima-media thickness based on neural networks and the shortest path method. The training process of the neural network model includes: Obtain a training dataset containing vascular ultrasound images and corresponding manually annotated lumen-intima and media-adventitia boundaries; After preprocessing the images in the training dataset, they are input into the initial neural network model for training. By optimizing the network parameters, the boundary probability map output by the model is made to approximate the true annotation, thus obtaining the trained neural network model.

[0033] Example 3: Based on Example 1, this example provides a method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method, including: In the step of dynamically assigning weights to edges in the graph based on the boundary probability map and the orientation field map, the weight W of the edge between the current pixel and its neighboring pixels is calculated using the following formula: ; in, This represents the probability value of the corresponding pixel in the boundary probability map; This represents the difference between the direction value of the current pixel in the direction field map and the direction vector pointing to neighboring pixels; This represents the auxiliary cost term calculated based on the image grayscale gradient. , , These are the weighting coefficients.

[0034] In this embodiment, This represents the probability value of the corresponding pixel in the boundary probability map (the probability value is between 0 and 1). The higher the probability value, the greater the possibility that this is a boundary. This design reverses the concept of "probability" to "cost," bridging neural networks and shortest path algorithms. The neural network output provides pixel-level classification probabilities; higher values ​​indicate a greater likelihood that the point belongs to the target boundary. The shortest path algorithm finds the path with the minimum cumulative cost; therefore, the boundaries the algorithm hopes to traverse are set to low-cost boundaries. In other words... The probability graph output by the neural network was transformed into a cost graph suitable for the shortest path method.

[0035] In this embodiment, α, β, and γ are weighting coefficients, which are initial values ​​set based on experience, with values ​​of 0.6, 0.3, and 0.1 respectively. These values ​​can be adjusted and optimized after multiple experiments. In this embodiment, It is the difference between the direction of the edge and the direction field of the blood vessel predicted by the neural network. The smaller the difference, the lower the weight, to ensure that the found boundary is smooth and consistent with the direction of the blood vessel.

[0036] Example 4: Based on Example 3, this example provides a method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method, including: Calculate the Using the following formula: ; in, This represents the value of the current pixel in the orientation field map; It represents the direction vector from the current center pixel to one of its neighboring pixels.

[0037] In this embodiment, This refers to the direction vector from the current center pixel to one of its neighboring pixels. Here, an 8-neighborhood is used, meaning each pixel is connected to pixels above, below, left, right, and in its four diagonal directions. With the current pixel `current` as the origin (0,0), its relative position to its neighboring pixels `neighbor` can be one of the following: (0,1), (1,0), (1,1), (0,-1), (-1,0), (-1,-1), (1,-1), (-1,1). It is the angle calculated from this vector.

[0038] Example 5: Based on Example 3, this example provides a method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method, including: Calculate the Using the following formula: ; in, This represents the gradient magnitude of neighboring pixels calculated using the image gradient operator on the original grayscale image.

[0039] In this embodiment, It uses the traditional image grayscale gradient as auxiliary information, with its weight reduced in areas of large gradient values. It is the image grayscale gradient magnitude calculated with neighboring pixels. It is obtained by convolving the original grayscale image using the Sobel gradient operator. The larger the value, the more likely the neighboring pixel is located near an image edge. In IMT measurements, the LII and MA boundaries of blood vessels are represented as edges in the image, i.e., regions with high gradient values.

[0040] In this embodiment, Similarly, the larger the gradient value, the smaller the cost. Utilizing the edge information of the image itself to supplement details that the neural network might miss makes the final cost function more robust.

[0041] Example 6: Based on Example 1, this example provides a method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method. The method uses the shortest path algorithm to search for the vascular boundary path, which includes: The Dijkstra algorithm is used to search for the minimum cost paths corresponding to the lumen-inner membrane boundary and the middle membrane-outer membrane boundary in the weighted graph.

[0042] In this embodiment, Dijkstra's algorithm is used to search for the minimum cost paths to the LII (luminal-intima boundary) and MA (media-adventitia boundary) boundaries in the graph to locate the vessel boundaries. Since the algorithm searches on an 8-connected graph, the paths are continuous.

[0043] In this embodiment, the boundary search based on the shortest path is as follows: Figure 4 As shown.

[0044] Example 7: Based on Example 1, this example provides a method for measuring intima-media thickness based on neural networks and the shortest path method. In the weighted graph, a start and end point are set for the vessel boundary to be searched, including: Based on the lumen-intima boundary and the media-outer membrane boundary, in the boundary probability map, the node with the highest corresponding boundary probability value in the leftmost column of the ultrasound image is selected as the starting point of the corresponding weighted map, and the node with the highest corresponding boundary probability value in the rightmost column of the ultrasound image is selected as the ending point of the weighted map.

[0045] Each of the two boundaries, LII (luminal-inner membrane boundary) and MA (middle-outer membrane boundary), has a pair of start and end points.

[0046] Example 8: Based on Example 1, this example provides a method for measuring intimal-media thickness based on neural networks and the shortest path method, calculating the average distance between the lumen-intima boundary and the media-adventitia boundary, including: Obtain each pixel node on the path between the middle and outer membranes, and calculate the shortest vertical distance from each pixel node to the path between the lumen and the inner membrane. Calculate the average of the shortest vertical distances of all pixel nodes as the average distance between the lumen-inner membrane boundary and the middle and outer membrane boundary.

[0047] Example 9: Based on Example 1, this example provides a method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method. The graph structure is constructed using pixels of vascular ultrasound images as nodes and the adjacency relationships between pixels as edges, including: Each pixel node is connected to the pixel nodes in its eight-neighborhood to form an edge; wherein the eight-neighborhood includes adjacent pixel nodes in the top, bottom, left, right and four diagonal directions.

[0048] Example 10: This example provides a system for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method, including: The image acquisition and preprocessing module is used to acquire and preprocess the ultrasound images of the blood vessels to be measured. The neural network feature extraction module is used to input the preprocessed vascular ultrasound image into the trained neural network model to obtain the boundary probability map and orientation field map output by the neural network model. The weighted graph construction module is used to construct a graph structure with pixels of vascular ultrasound images as nodes and the adjacency relationship between pixels as edges. It dynamically assigns weights to the edges in the graph based on the boundary probability map and the orientation field map to obtain a weighted graph. The boundary path search module is used to set the start and end points of the blood vessel boundary to be searched in the weighted graph, and to search for the blood vessel boundary path based on the shortest path algorithm. The thickness calculation and output module is used to calculate the average distance between the lumen-intima boundary and the media-adventitia boundary based on the vascular boundary path, and output the vascular intima-media thickness measurement results.

[0049] The beneficial effects of the above technical solution are: higher accuracy and robustness; the neural network can adaptively learn complex image features and contextual information, automatically identify vascular features, overcome the limitations of manually designed cost functions, and perform more stably and accurately under complex conditions such as low image quality, plaques, and noise interference; improved performance; leveraging the advantages of neural networks in feature learning and the advantages of shortest path algorithms in structural optimization; high degree of automation; fully automatic and high-precision IMT measurement without extensive manual intervention, with good repeatability; potential for portable devices; through lightweight model design and optimization, this method can be deployed in handheld ultrasound devices with limited computing power to achieve rapid screening at the bedside or in the community.

[0050] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method, characterized in that, include: Acquire ultrasound images of the blood vessels to be measured and perform preprocessing; The preprocessed vascular ultrasound image is input into the trained neural network model to obtain the boundary probability map and orientation field map output by the neural network model. A graph structure is constructed using pixels of vascular ultrasound images as nodes and adjacency relationships between pixels as edges. Weights are dynamically assigned to the edges in the graph based on boundary probability maps and orientation field maps to obtain a weighted graph. In the weighted graph, a start point and an end point are set for the blood vessel boundary to be searched, and the blood vessel boundary path is obtained based on the shortest path algorithm; Based on the vascular boundary path, the average distance between the lumen-intima boundary and the media-adventitia boundary is calculated, and the intima-media thickness measurement results are output.

2. The method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method according to claim 1, characterized in that, The training process of the neural network model includes: Obtain a training dataset containing vascular ultrasound images and corresponding manually annotated lumen-intima and media-adventitia boundaries; After preprocessing the images in the training dataset, they are input into the initial neural network model for training. By optimizing the network parameters, the boundary probability map output by the model is made to approximate the true annotation, thus obtaining the trained neural network model.

3. The method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method according to claim 1, characterized in that, include: In the step of dynamically assigning weights to edges in the graph based on the boundary probability map and the orientation field map, the weight W of the edge between the current pixel and its neighboring pixels is calculated using the following formula: ； in, This represents the probability value of the corresponding pixel in the boundary probability map; Indicates the boundary; This represents the difference between the direction value of the current pixel in the direction field map and the direction vector pointing to neighboring pixels; This represents the auxiliary cost term calculated based on the image grayscale gradient. , , These are the weighting coefficients.

4. The method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method according to claim 3, characterized in that, include: Calculate the Using the following formula: ； in, This represents the value of the current pixel in the orientation field map; It represents the direction vector from the current center pixel to one of its neighboring pixels.

5. The method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method according to claim 3, characterized in that, include: Calculate the Using the following formula: ； in, This represents the gradient magnitude of neighboring pixels calculated using the image gradient operator on the original grayscale image.

6. The method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method according to claim 1, characterized in that, The blood vessel boundary paths obtained based on the shortest path algorithm include: The Dijkstra algorithm is used to search for the minimum cost paths corresponding to the lumen-inner membrane boundary and the middle membrane-outer membrane boundary in the weighted graph.

7. The method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method according to claim 1, characterized in that, In the weighted graph, a start point and an end point are defined for the boundary of the blood vessel to be searched, including: Based on the lumen-intima boundary and the media-outer membrane boundary, in the boundary probability diagram, the node with the highest corresponding boundary probability value in the leftmost column of the ultrasound image is selected as the starting point of the corresponding weighted diagram, and the node with the highest corresponding boundary probability value in the rightmost column of the ultrasound image is selected as the ending point of the weighted diagram.

8. The method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method according to claim 1, characterized in that, Calculate the average distance between the lumen-inner membrane boundary and the middle-outer membrane boundary, including: Obtain each pixel node on the path between the middle and outer membranes, and calculate the shortest vertical distance from each pixel node to the path between the lumen and the inner membrane. Calculate the average of the shortest vertical distances of all pixel nodes as the average distance between the lumen-inner membrane boundary and the middle and outer membrane boundary.

9. The method for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method according to claim 1, characterized in that, A graph structure is constructed using pixels from vascular ultrasound images as nodes and adjacency relationships between pixels as edges, including: Each pixel node is connected to the pixel nodes in its eight-neighborhood to form an edge; wherein the eight-neighborhood includes adjacent pixel nodes in the top, bottom, left, right and four diagonal directions.

10. A system for measuring the thickness of the vascular intima-media layer based on neural networks and the shortest path method, characterized in that, include: The image acquisition and preprocessing module is used to acquire and preprocess the ultrasound images of the blood vessels to be measured. The neural network feature extraction module is used to input the preprocessed vascular ultrasound image into the trained neural network model to obtain the boundary probability map and orientation field map output by the neural network model. The weighted graph construction module is used to construct a graph structure with pixels of vascular ultrasound images as nodes and the adjacency relationship between pixels as edges. It dynamically assigns weights to the edges in the graph based on the boundary probability map and the orientation field map to obtain a weighted graph. The boundary path search module is used to set the start and end points of the blood vessel boundary to be searched in the weighted graph, and to search for the blood vessel boundary path based on the shortest path algorithm. The thickness calculation and output module is used to calculate the average distance between the lumen-intima boundary and the media-adventitia boundary based on the vascular boundary path, and output the vascular intima-media thickness measurement results.

Images