Image Processing Method and Related Devices, Electronic Equipment, and Storage Medium

Abstract

The present application discloses an image processing method and related apparatuses, an electronic device, and a storage medium. Among them, the image processing method includes: performing contour detection on a to-be-detected blood vessel image to obtain a first image region of the blood vessel; identifying the blood vessel category to which each pixel point in the first image region belongs, and extracting a plurality of first pixel points representing the blood vessel path in the first image region; wherein, the blood vessel category includes a plurality of preset categories; based on the reference pixel points belonging to the reference category, respectively performing path analysis on the first pixel points belonging to the same blood vessel category to obtain the blood vessel paths of each blood vessel category; wherein, the blood vessels of the reference category are in a preset position among the blood vessels of the plurality of preset categories; based on the blood vessel paths, optimizing the first image region to obtain a target blood vessel region belonging to the target category. The above solution can improve the detection efficiency of the blood vessel region.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to an image processing method, related devices, electronic devices, and storage media. Background Art

[0002] Vascular images are of extremely important significance in scenarios such as assisting doctors in surgical planning. For example, in head and neck CTA (Computed Tomography Angiography), through a series of operations, CPR (Curved-Planar Reconstruction) images of the internal carotid artery, vertebral artery, etc. can be obtained to provide doctors with rich information related to the head and neck region.

[0003] Currently, doctors still need to manually analyze and process vascular images to obtain the final vascular region. However, this process is usually very cumbersome and time-consuming, thus affecting the doctor's work efficiency. In view of this, how to improve the detection efficiency of the vascular region has become an urgent problem to be solved. Summary of the Invention

[0004] This application provides an image processing method, related devices, electronic devices, and storage media.

[0005] In a first aspect of this application, an image processing method is provided, including: performing contour detection on a to-be-detected vascular image to obtain a first image region of the blood vessel; identifying the blood vessel category to which each pixel point in the first image region belongs, and extracting a plurality of first pixel points representing the blood vessel path in the first image region; where the blood vessel category includes a plurality of preset categories; based on reference pixel points belonging to a reference category, respectively performing path analysis on the first pixel points belonging to the same blood vessel category to obtain the blood vessel paths of each blood vessel category; where the blood vessels of the reference category are in a preset position among the blood vessels of the plurality of preset categories; based on the blood vessel path, optimizing the first image region to obtain a target blood vessel region belonging to a target category.

[0006] Therefore, perform contour detection on the blood vessel image to be measured, obtain the first image region of the blood vessel, identify the blood vessel category to which each pixel point in the first image region belongs, and extract a number of first pixel points representing the blood vessel path in the first image region. The blood vessel category includes a number of preset categories. On this basis, based on the reference pixel points belonging to the reference category, analyze the paths of the first pixel points belonging to the same blood vessel category respectively to obtain the blood vessel paths of each blood vessel category. The blood vessel of the reference category is in a preset position among the blood vessels of a number of preset categories. And based on the blood vessel path, optimize the first image region to obtain the target blood vessel region belonging to the target category. On the one hand, since there is no need to manually process and analyze the image to obtain the blood vessel region, it is beneficial to improve the image detection efficiency. On the other hand, in the process of path analysis, by using the reference pixel points subordinate to the blood vessel of the reference category that is in a preset position among the blood vessels of a number of preset categories to analyze the first pixel points belonging to the same blood vessel category to obtain the blood vessel path of this blood vessel category, it is beneficial to eliminate interference as much as possible and improve the accuracy of image detection.

[0007] Among them, based on the reference pixel points belonging to the reference category, analyze the paths of the first pixel points belonging to the same blood vessel category respectively to obtain the blood vessel paths of each blood vessel category, including: taking each blood vessel category as the current category respectively; determining the end point of the blood vessel path of the current category based on the distances from each first pixel point belonging to the current category to the reference pixel points; and sequentially connecting each first pixel point belonging to the current category based on the end point of the blood vessel path to obtain the blood vessel path of the current category.

[0008] Therefore, taking each blood vessel category as the current category respectively, determining the end point of the blood vessel path of the current category based on the distances from each first pixel point belonging to the current category to the reference pixel points, and sequentially connecting each first pixel point belonging to the current category based on the end point of the blood vessel path to obtain the blood vessel path of the current category, that is, in the process of path analysis, the end point of the blood vessel path can be determined first, and then each first pixel point is connected according to the end point of the blood vessel path to obtain the blood vessel path, which can avoid the influence of other categories on the path analysis of the current category as much as possible and is beneficial to improving the detection speed.

[0009] Among them, the preset position is the center position, and the end point of the blood vessel path is the first pixel point with the maximum distance; and / or, each first pixel point belonging to the current category is sequentially connected by using the shortest path algorithm.

[0010] Therefore, setting the preset position as the center position and setting the end point of the blood vessel path as the first pixel point with the maximum distance can greatly reduce the complexity of determining the end point of the blood vessel path and is beneficial to further improving the detection speed; while sequentially connecting each first pixel point belonging to the current category by using the shortest path algorithm to obtain the blood vessel path can greatly reduce the probability of error in connecting the first pixel points and improve the accuracy of the blood vessel path.

[0011] Among them, after sequentially connecting each first pixel point belonging to the current category based on the end point of the blood vessel path to obtain the blood vessel path of the current category, the method further includes: checking whether there is a blood vessel path of the same blood vessel category; in response to the existence of a blood vessel path of the same blood vessel category, connecting the blood vessel paths of the same blood vessel category.

[0012] Therefore, after obtaining the blood vessel path of the current category, further checking whether there is a blood vessel path of the same blood vessel category and, in response to the existence of a blood vessel path of the same blood vessel category, connecting the blood vessel paths of the same blood vessel category can greatly alleviate the problem of blood vessel image breakage caused by reasons such as blood vessel stenosis and angiography, which is beneficial to improving the detection accuracy.

[0013] Among them, optimizing the first image region based on the blood vessel path to obtain the target blood vessel region belonging to the target category includes: cropping the blood vessel regions that do not belong to the target category based on the blood vessel paths of each blood vessel category to obtain a second image region; for each pixel point in the second image region, based on the deviation between the first distance corresponding to the pixel point and the second distance corresponding to the pixel point, removing the pixel points whose deviation meets the preset conditions to obtain the target blood vessel region belonging to the target category; where the first distance represents the shortest distance from the pixel point to each first pixel point, and the second distance represents the shortest distance from the pixel point to the target path.

[0014] Therefore, cropping the blood vessel regions that do not belong to the target category based on the blood vessel paths of each blood vessel category to obtain a second image region, and on this basis, for each pixel point in the second image region, based on the deviation between the first distance corresponding to the pixel point and the second distance corresponding to the pixel point, removing the pixel points whose deviation meets the preset conditions to obtain the target blood vessel region belonging to the target category can, on the one hand, cut off the blood vessel regions of irrelevant categories, and on the other hand, by measuring the deviation between the first distance and the second distance and removing the pixel points whose deviation meets the preset conditions, can further optimize the blood vessel region, which is beneficial to improving the accuracy of the blood vessel region.

[0015] Among them, the blood vessel path is represented by the blood vessel center line, and the image processing method includes: respectively taking each first pixel point on the target path as the current pixel point; obtaining the truncation reference of the normal reference of the target path at the current pixel point in the second image region; taking the pixel point located at the center position of the truncation reference as the second pixel point corresponding to the current pixel point; connecting the second pixel points corresponding to each first pixel point to obtain the optimized target path; where, when the blood vessel image to be measured is a three-dimensional image, the normal reference is a normal plane and the truncation reference is a truncation plane, and when the blood vessel image to be measured is a two-dimensional image, the normal reference is a normal line and the truncation reference is a truncation line.

[0016] Therefore, the blood vessel path is represented by the blood vessel center line. Before measuring the second distance, each first pixel point on the target path is used as the current pixel point, and the truncation reference of the normal reference of the target path at the current pixel point in the second image region is obtained. And the pixel point located at the center position of the truncation reference is used as the second pixel point corresponding to the current pixel point. Based on this, the second pixel points corresponding to each first pixel point are connected to optimize the target path. And when the blood vessel image to be measured is a three-dimensional image, the normal reference is a normal plane and the truncation reference is a truncation plane. When the blood vessel image to be measured is a two-dimensional image, the normal reference is a normal line and the truncation reference is a truncation line. That is, before measuring the second distance, further optimizing the target path can improve the accuracy of the target path and is beneficial to the accuracy of the finally optimized blood vessel region.

[0017] Wherein, the deviation is the absolute value of the difference between the first distance and the second distance, and the preset condition includes that the deviation is greater than the distance threshold; and / or, the deviation corresponding to the pixel point is calculated in parallel by the GPU.

[0018] Therefore, the deviation is set as the absolute value of the difference between the first distance and the second distance, and the preset condition is set to include that the deviation is greater than the distance threshold. Therefore, pixel points with too large deviations can be eliminated, which is beneficial to reducing the complexity of eliminating pixel points. And calculating the deviation corresponding to each pixel point in parallel by the GPU can improve the overall calculation speed of the deviation, thus being beneficial to improving the detection efficiency.

[0019] Wherein, identifying the blood vessel category to which each pixel point in the first image region belongs includes: obtaining first image data based on the pixel values of each pixel point in the blood vessel image to be measured respectively, and obtaining second image data based on the offset values between each pixel point and the reference position respectively; wherein, the first image data includes the pixel values of each pixel point, and the second image data includes the offset values corresponding to each pixel point; fusing the first image data and the second image data to obtain fused image data; and using an identification model to identify the fused image data to obtain the blood vessel category to which each pixel point belongs.

[0020] Therefore, first image data is obtained based on the pixel values of each pixel point in the blood vessel image to be measured respectively, and second image data is obtained based on the offset values between each pixel point and the reference position respectively. And the first image data includes the pixel values of each pixel point, and the second image data includes the offset values corresponding to each pixel point. On this basis, the first image data and the second image data are fused to obtain fused image data, and an identification model is used to identify the fused image data to obtain the blood vessel category to which each pixel point belongs. Therefore, in the process of category identification, on the one hand, the texture feature information represented by the first image data is referred to, and on the other hand, the geometric feature information represented by the second image data can be referred to, which is beneficial to improving the accuracy of blood vessel category identification.

[0021] Among them, before using the recognition model to recognize the fused image data to obtain the blood vessel category to which each pixel belongs, the method further includes: extracting a plurality of sub-image data from the fused image data; wherein, the set of the plurality of sub-image data covers the first image area; using the recognition model to recognize the fused image data to obtain the blood vessel category to which each pixel belongs, including: using the recognition model to recognize the plurality of sub-image data respectively to obtain the category recognition results corresponding to the respective sub-image data; wherein, the category recognition results include the probability values that the pixel belongs to a plurality of preset categories respectively; based on the category recognition results corresponding to the respective sub-image data, obtaining the blood vessel category to which each pixel belongs.

[0022] Therefore, before using the recognition model to recognize the fused image data, a plurality of sub-image data are first extracted from the fused image data, and the set of the plurality of sub-image data can cover the first image area. On this basis, the recognition model is used to recognize the plurality of sub-image data respectively to obtain the category recognition results corresponding to the respective sub-image data, and the category recognition results include the probability values that the pixel belongs to a plurality of preset categories respectively. Then, based on the category recognition results corresponding to the respective sub-image data, the blood vessel category to which each pixel belongs is obtained. Therefore, by converting the overall recognition of the fused image data into the sub-graph recognition of the respective sub-image data, it is beneficial to greatly reduce the computing load and improve the detection speed.

[0023] Among them, the plurality of preset categories include: aorta, left external carotid artery, right external carotid artery, left subclavian artery, right subclavian artery, left internal carotid artery, right internal carotid artery, left vertebral artery, right vertebral artery. The reference category includes the aorta, and the target categories include: left internal carotid artery, right internal carotid artery, left vertebral artery, right vertebral artery.

[0024] Therefore, the plurality of preset categories are set to include: aorta, left external carotid artery, right external carotid artery, left subclavian artery, right subclavian artery, left internal carotid artery, right internal carotid artery, left vertebral artery, right vertebral artery, the reference category is set to include the aorta, and the target categories are set to include: left internal carotid artery, right internal carotid artery, left vertebral artery, right vertebral artery. That is, it can be applied to the blood vessel detection process of the head and neck arteries, and detect and obtain the blood vessel regions of the internal carotid artery and the vertebral artery, which is beneficial to provide rich auxiliary information in the head and neck parts.

[0025] The second aspect of the present application provides an image processing device, including: a contour detection module, a category recognition module, a path analysis module, and a region optimization module. The contour detection module is configured to perform contour detection on a to-be-tested blood vessel image to obtain a first image region of the blood vessel. The category recognition module is configured to identify the blood vessel category to which each pixel point in the first image region belongs, and extract a plurality of first pixel points representing the blood vessel path in the first image region; wherein, the blood vessel category includes a plurality of preset categories. The path analysis module is configured to perform path analysis on the first pixel points belonging to the same blood vessel category respectively based on the reference pixel points belonging to the reference category to obtain the blood vessel paths of each blood vessel category; wherein, the blood vessels of the reference category are in a preset position among the blood vessels of the plurality of preset categories. The region optimization module is configured to optimize the first image region based on the blood vessel path to obtain a target blood vessel region belonging to the target category.

[0026] The third aspect of the present application provides an electronic device, including a memory and a processor coupled to each other. The processor is configured to execute program instructions stored in the memory to implement the image processing method in the first aspect above.

[0027] The fourth aspect of the present application provides a computer-readable storage medium, on which program instructions are stored. When the program instructions are executed by a processor, the image processing method in the first aspect above is implemented.

[0028] In the above solution, contour detection is performed on the to-be-tested blood vessel image to obtain a first image region of the blood vessel, and the blood vessel category to which each pixel point in the first image region belongs is identified, and a plurality of first pixel points representing the blood vessel path are extracted in the first image region, and the blood vessel category includes a plurality of preset categories. On this basis, path analysis is performed on the first pixel points belonging to the same blood vessel category respectively based on the reference pixel points belonging to the reference category to obtain the blood vessel paths of each blood vessel category, and the blood vessels of the reference category are in a preset position among the blood vessels of the plurality of preset categories, and the first image region is optimized based on the blood vessel path to obtain a target blood vessel region belonging to the target category. On the one hand, since there is no need to manually process and analyze the image to obtain the blood vessel region, it is beneficial to improve the image detection efficiency. On the other hand, in the process of path analysis, by using the reference pixel points under the reference category of blood vessels that are in a preset position among the blood vessels of the plurality of preset categories to analyze the first pixel points belonging to the same blood vessel category to obtain the blood vessel path of this blood vessel category, it is beneficial to exclude interference as much as possible and improve the accuracy of image detection. Description of the Drawings

[0029] Figure 1 is a schematic flowchart of an embodiment of the image processing method of the present application;

[0030] Figure 2 A schematic framework diagram of an embodiment of the recognition model;

[0031] Figure 3 It is a schematic framework diagram of an embodiment of the image processing apparatus of the present application;

[0032] Figure 4 It is a schematic framework diagram of an embodiment of the electronic device of the present application;

[0033] Figure 5 It is a schematic framework diagram of an embodiment of the computer-readable storage medium of the present application. Detailed implementation manners

[0034] Next, in conjunction with the accompanying drawings of the specification, the solutions of the embodiments of the present application will be described in detail.

[0035] In the following description, specific details such as specific system architectures, interfaces, and technologies are proposed for the purpose of illustration rather than limitation, so as to thoroughly understand the present application.

[0036] The terms "system" and "network" are often used interchangeably in this article. The term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this article generally represents an "or" relationship between the front and rear associated objects. In addition, "multiple" in this article means two or more than two.

[0037] Please refer to Figure 1 , Figure 1 It is a schematic flowchart of an embodiment of the image processing method of the present application.

[0038] Specifically, it may include the following steps:

[0039] Step S11: Perform contour detection on the blood vessel image to be measured to obtain the first image area of the blood vessel.

[0040] In one implementation scenario, in order to improve the efficiency of contour detection, a contour detection model can be pre-trained, based on which, the contour detection model can be used to perform contour detection on the blood vessel image to be detected to obtain the first image area of the blood vessel. The contour detection model can specifically include but is not limited to U-Net, etc., and the network structure of the contour detection model is not limited here. It should be noted that in the process of training the contour detection model, a number of sample blood vessel images can be collected in advance, and the sample blood vessel images are annotated with sample image areas belonging to blood vessels. Based on this, the contour detection model can be used to perform contour detection on the sample blood vessel images to obtain sample detection results of whether each sample pixel point in the sample blood vessel image belongs to a blood vessel. On this basis, the connected domain formed by the sample pixel points belonging to the blood vessel can be used as the predicted blood vessel area, so that the network parameters of the contour detection network can be adjusted based on the difference between the sample blood vessel area and the predicted blood vessel area. Specifically, the difference between the sample blood vessel area and the predicted blood vessel area can be measured based on loss functions such as cross entropy loss function and dice loss, and the network parameters can be adjusted based on optimization methods such as gradient descent. The specific process can refer to the technical details of the above-mentioned loss function and optimization method, which will not be repeated here. In addition, after the contour detection model is trained, the contour detection can be used to detect the blood vessel image to be tested to obtain the detection result of whether each pixel point in the blood vessel image to be tested belongs to the blood vessel. On this basis, the connected domain formed by the pixel points belonging to the blood vessel can be used as the first image area of the blood vessel.

[0041] In one implementation scenario, in order to improve the image contrast and eliminate the interference of irrelevant areas on contour detection as much as possible to improve the accuracy of contour detection, before contour detection, the vascular image to be tested may be preprocessed, and the preprocessing operation may specifically include but is not limited to: windowing, normalization, etc., which are not limited here. Taking the vascular image to be tested as a CT (Computed Tomography) image as an example, the vascular image to be tested may be windowed using a window function with a preset window width and a preset window position, so that the pixel values in the vascular image to be tested that are located outside the boundary of the window function are reset to the boundary value of the window function. Exemplarily, the window function may be set to [-100, 700], and the pixel points with HU (i.e., Houns field unit) values lower than -100 in the vascular image to be tested may be reset to -100, and the pixel points with HU values higher than 700 in the vascular image to be tested may be reset to 700. Other situations may be deduced by analogy, and examples are not given one by one here. In addition, after the windowing process, the pixel value of each pixel point in the windowed blood vessel image to be measured may be normalized, that is, the pixel value of each pixel point is normalized to a value range of 0 to 1.

[0042] Step S12: Identify the blood vessel category to which each pixel point in the first image region belongs, and extract a number of first pixel points representing the blood vessel path in the first image region.

[0043] In the embodiments of the present disclosure, the blood vessel categories include a number of preset categories, which can be specifically set according to actual application needs. Taking the head and neck region as an example, the number of preset categories may include, but is not limited to: aorta, left external carotid artery, right external carotid artery, left subclavian artery, right subclavian artery, left internal carotid artery, right internal carotid artery, left vertebral artery, right vertebral artery, etc., which are not limited herein. The same can be applied to other parts, and no further examples will be given here.

[0044] In an implementation scenario, in order to improve the category recognition efficiency, a recognition model can be pre-trained, and the recognition model is used to identify the blood vessel category to which each pixel point in the first image region belongs. The recognition model may specifically include, but is not limited to, a convolutional neural network, etc., and the network structure of the recognition model is not specifically limited herein. Specifically, a number of sample images can be pre-collected, and the sample categories to which the sample pixel points belonging to the blood vessels in the sample images are labeled. On this basis, the prediction categories to which the sample pixel points belonging to the blood vessels in the sample images belong can be identified by using the recognition model. For example, the recognition model can output the prediction probability values that the sample pixel points belong to a number of preset categories respectively, and use the preset category corresponding to the maximum prediction probability value as the prediction category to which the sample pixel point belongs. On this basis, the network parameters of the recognition model can be adjusted based on the difference between the sample category and the prediction category. It should be noted that a loss function such as cross entropy can be used to measure the above difference, and an optimization method such as gradient descent can be used to adjust the above parameters. The specific process can refer to the technical details of the above loss function and optimization method, which will not be elaborated herein. In addition, after the recognition model is trained, the probability values that the pixel points in the first image region belong to a number of preset categories can be identified by using the recognition model, and the preset category corresponding to the maximum probability value is used as the blood vessel category to which the pixel point belongs.

[0045] In an implementation scenario, in order to improve the efficiency of class recognition while further enhancing the recognition accuracy, first image data can be obtained based on the pixel values of each pixel point in the first image region in the blood vessel image to be measured, and second image data can be obtained based on the offset values between each pixel point and the reference position. The first image data includes the pixel values of each pixel point, and the second image data includes the offset values corresponding to each pixel point. On this basis, the first image data and the second image data are fused to obtain fused image data, and the recognition model is used to recognize the fused image data to obtain the blood vessel class to which each pixel point belongs. In the above manner, during the class recognition process, on the one hand, the texture feature information represented by the first image data is referred to, and on the other hand, the geometric feature information represented by the second image data can be referred to, which is beneficial to improving the accuracy of blood vessel class recognition.

[0046] In a specific implementation scenario, taking the blood vessel image to be measured as a three-dimensional image as an example, a matrix with the size of C*W*D*H can be used to store the pixel values of each pixel point as the first image data, where C represents the number of channels, W represents the width, D represents the depth, and H represents the height. For example, as described above, the blood vessel image to be measured can be windowed and normalized, and the pixel values of each pixel point belonging to the blood vessel are represented by one-dimensional numerical values within the range of 0 to 1. Therefore, the number of channels C is 1. In addition, the width W can be the width of the blood vessel image to be measured, the depth D can be the depth of the blood vessel image to be measured, and the height H can be the height of the blood vessel image to be measured. For example, when the width, depth, and height of the blood vessel image to be measured are all 1024, a matrix of 1*1024*1024*1024 can be used to store the pixel values of each pixel point as the first image data. Other cases can be deduced by analogy, and no further examples will be given here.

[0047] In a specific implementation scenario, still taking the blood vessel image to be measured as a three-dimensional image as an example, a matrix with the size of C*W*D*H can be used to store the offset values of each pixel point as the second image data, where C represents the number of channels, W represents the width, D represents the depth, and H represents the height. In addition, the reference position can be specifically set as the centroid of the first image region, that is, the blood vessel centroid, which is not limited here. On this basis, the offset value can be calculated based on the Euclidean distance. At this time, the number of channels C is 1. Of course, the offset value can be represented by the offsets in three directions: width, depth, and height. At this time, the number of channels C is 3, which is not limited here. In addition, the specific meanings of the width W, depth D, and height H can refer to the relevant descriptions above and will not be elaborated here. Still taking the width, depth, and height of the blood vessel image to be measured as 1024 and the number of channels C as 1 as an example, a matrix of 1*1024*1024*1024 can be used to store the offset values of each pixel point as the second image data. Other cases can be deduced by analogy, and no further examples will be given here.

[0048] In a specific implementation scenario, the first image data and the second image data can be combined to obtain fused image data. For example, as described above, a matrix with dimensions C*W*D*H can be used to store the pixel values of each pixel point as the first image data, and a matrix with dimensions C*W*D*H can also be used to store the offset values of each pixel point as the second image data. Then the fused image data can be represented as 2C*W*D*H, that is, each pixel point has two attribute values, namely the pixel value and the offset value.

[0049] In a specific implementation scenario, by using an identification model to identify the fused image data, the probability values that the pixel points belong to several preset categories can be obtained. On this basis, the preset category corresponding to the maximum probability value can be used as the blood vessel category to which the pixel point belongs. Still taking the head and neck area as an example, through the identification of the identification model, the probability values that the pixel points belong to the aorta, left external carotid artery, right external carotid artery, left subclavian artery, right subclavian artery, left internal carotid artery, right internal carotid artery, left vertebral artery, and right vertebral artery can be obtained. On this basis, the preset category corresponding to the maximum probability value can be used as the blood vessel category to which the pixel point belongs.

[0050] In a specific implementation scenario, during the training process, several sample images can be collected in advance, and each sample pixel point in the sample image is labeled with a sample label, and the sample label indicates the blood vessel category to which the sample pixel point belongs. Exemplarily, taking the head and neck area as an example, the above nine preset categories can be represented by the numbers 1 to 9 respectively, and 0 can be used to label the background. On this basis, based on the pixel values of the sample pixel points belonging to the blood vessels in the sample image, the first sample image data can be obtained, and based on the offset values between these sample pixel points and the reference position, the second sample image data can be obtained. The first sample image data includes the pixel values of these sample pixel points, and the second sample image data includes the offset values of these sample pixel points. Based on this, the first sample image data and the second sample image data can be fused to obtain the fused sample image data, and the identification model can be used to identify the fused sample image data to obtain the blood vessel categories to which these sample pixel points belong. Thus, based on the difference between the blood vessel categories labeled by the sample pixel points and the predicted blood vessel categories, the network parameters of the identification model can be adjusted. The specific measurement method of the difference and the specific adjustment method of the parameters can refer to the relevant descriptions above and will not be elaborated here.

[0051] In a specific implementation scenario, in order to reduce the video memory occupancy of the identification model during the identification process, the identification model can be specifically set to HRNET. Please refer to Figure 2 , Figure 2 which is a schematic framework diagram of an embodiment of the identification model. Specifically, Figure 2 is a schematic framework diagram of an embodiment of HRNET. AsFigure 2 As shown, each layer represents a subnet of a different size. Starting from the high-resolution subnet in the first stage, the subnets from high resolution to low resolution are gradually increased, which can form more stages, and the subnets of each resolution are connected in parallel. In addition, information is repeatedly exchanged between subnets through operations such as upsampling and downsampling, which can promote multi-scale feature fusion. For the specific data processing process of HRNET, refer to the technical details of HRNET, which will not be elaborated here.

[0052] In a specific implementation scenario, in order to further reduce the video memory occupation of the recognition model during the recognition process, before using the recognition model for recognition, several sub-image data can also be extracted from the fused image data, and the set of several sub-image data can cover the first image area, that is to say, the set of several sub-image data can cover the entire blood vessel. For example, pixel points can be randomly selected in the first image area as sampling pixel points, and the area data centered on the sampling pixel points and with a size of a preset size (such as 32*32*32) in the fused image data can be extracted as sub-image data. On this basis, the recognition model can be used to recognize several sub-image data respectively to obtain the category recognition results corresponding to each sub-image data, and the category recognition results include the probability values that the pixel points belong to several preset categories respectively, and based on the category recognition results corresponding to each sub-image data, the blood vessel categories to which each pixel point belongs can be obtained. Exemplarily, if a pixel point only exists in one sub-image data, its blood vessel category to which it belongs can be determined based on the maximum value among the probability values that the pixel point belongs to several preset categories respectively. For example, the preset category corresponding to the maximum probability value can be directly used as the blood vessel category to which the pixel point belongs; or, if a pixel point exists in multiple sub-image data, that is, there is an overlapping part among the sub-image data, the pixel points in the overlapping part will exist in multiple sub-image data. At this time, there are multiple probability values corresponding to each preset category for the pixel point. Then, for each preset category, the average value of the probability values that the pixel point belongs to this preset category can be taken as the final probability value that it belongs to this preset category, and the preset category corresponding to the maximum final probability value can be further used as the blood vessel category to which the pixel point belongs. The above method can convert the overall recognition of the fused image data into the sub-graph recognition of each sub-image data, which is beneficial to greatly reducing the computing load and improving the detection speed.

[0053] In an implementation scenario, the topological structure of the first image area can be established, and based on the topological structure, several first pixel points representing the blood vessel path can be extracted.

[0054] In a specific implementation scenario, starting from the inner edge pixel points of the first image region, a topological structure is gradually established. It should be noted that the edge pixel points refer to the pixel points located inside the first image region and adjacent to the background pixel points. On this basis, for each edge pixel point, the non-edge pixel point closest to it can be searched for inside the first image region and used as the leaf node of this non-edge pixel point. After determining the non-edge pixel point closest to each edge pixel point, the non-edge pixel points located inside the first image region and adjacent to the edge pixel points can be used as new edge pixel points, and for each new edge pixel point, the above steps of searching for the non-edge pixel point closest to it inside the first image region and using it as the leaf node of this non-edge pixel point are executed, and so on, until all the non-edge pixel points that have not been used as edge pixel points already have leaf nodes, then the topological structure of the first image region can be established.

[0055] In a specific implementation scenario, for this topological structure, the pixel points without leaf nodes can be deleted first to obtain a new topological structure. On this basis, on the basis of the new topological structure, the pixel points without leaf nodes are deleted again to obtain a new topological structure, and so on, until the remaining pixel points in the topological structure do not have leaf nodes, then these remaining pixel points are the first pixel points representing the blood vessel path.

[0056] Step S13: Based on the reference pixel points belonging to the reference category, path analysis is performed on the first pixel points belonging to the same blood vessel category respectively to obtain the blood vessel paths of each blood vessel category.

[0057] In the embodiments of the present disclosure, the blood vessels of the reference category are in a preset position among several preset categories of blood vessels. For example, the preset position can be set as the central position. Still taking the head and neck part as an example, as mentioned above, several preset categories include but are not limited to: aorta, left external carotid artery, right external carotid artery, left subclavian artery, right subclavian artery, left internal carotid artery, right internal carotid artery, left vertebral artery, right vertebral artery. Then, when the preset position is the central position, the reference category can be specifically set as the aorta. Other situations can be deduced by analogy and will not be exemplified one by one here.

[0058] In an implementation scenario, it should be noted that the reference pixel points are located in the blood vessels of the reference category. Specifically, among the pixel points belonging to the reference category in the first image region, the pixel points located at the target position can be used as the reference pixel points. Specifically, the target position can be set according to the actual application, such as it can be set as the starting position. Still taking the head and neck part as an example, as mentioned above, the reference category can be set as the aorta. Then, when the target position is set as the starting position, the reference pixel points represent the pixel points at the starting point of the aorta. Other situations can be deduced by analogy and will not be exemplified one by one here.

[0059] In one implementation scenario, each blood vessel category can be used as the current category, and based on the distances from each first pixel point belonging to the current category to the reference pixel point, the end point of the blood vessel path of the current category is determined. On this basis, based on the end point of the blood vessel path, each first pixel point belonging to the current category is sequentially connected to obtain the blood vessel path of the current category. In the above method, during the path analysis process, the end point of the blood vessel path can be determined first, and then the blood vessel path can be obtained by connecting each first pixel point according to the end point of the blood vessel path, thereby avoiding the influence of other categories on the path analysis of the current category as much as possible, which is conducive to improving the detection speed.

[0060] In a specific implementation scenario, as mentioned above, the preset position can be set to the center position, and the end point of the vascular path can be the first pixel point with the largest distance. That is to say, after calculating the distances from each first pixel point belonging to the current category to the reference pixel point, the first pixel point with the largest distance can be selected as the end point of the vascular path of the current category. Still taking the head and neck as an example, when the current category is the left internal carotid artery, the distances from each first pixel point belonging to the left internal carotid artery to the starting point of the aorta can be calculated, and the first pixel point with the largest distance can be selected as the end point of the vascular path of the left internal carotid artery. Other situations can be deduced by analogy, and examples are not given one by one here.

[0061] In a specific implementation scenario, the shortest path algorithm can be used to sequentially connect the first pixel points belonging to the current category to obtain the vascular path of the current category. That is, for each first pixel point belonging to the current category, the shortest path algorithm can be used to trace the end point of the vascular path of the current category to obtain the vascular path of the current category. It should be noted that the shortest path algorithm may include but is not limited to the Dijkstra algorithm, which is not limited here. For the specific connection process, please refer to the technical details of the shortest path algorithm such as the Dijkstra algorithm, which will not be repeated here.

[0062] In a specific implementation scenario, in order to further improve the accuracy of the vascular path, after sequentially connecting the first pixel points belonging to the current category to obtain the vascular path of the current category, it is also possible to check whether there is a vascular path of the same vascular category, and in response to the existence of a vascular path of the same vascular category, connect the vascular paths of the same vascular category. It should be noted that in real scenarios, vascular images may be broken due to vascular stenosis, angiography, etc., so the above method can connect the broken blood vessels to improve the accuracy of the vascular path. Specifically, when there are vascular paths of the same vascular category, the first pixel points closest to each other in the vascular paths of the same vascular category can be connected to obtain the final vascular path of the vascular category, which can greatly alleviate the problem of vascular image breakage.

[0063] Step S14: Optimize the first image region based on the vascular path to obtain a target vascular region belonging to the target category.

[0064] In one implementation scenario, the target category is at least one of the above-mentioned several preset categories. Specifically, at least one of the above-mentioned several preset categories can be specified as the target category. For example, the target category can be default-specified or specified by the user. Still taking the head and neck region as an example, as mentioned above, the several preset categories can include, but are not limited to: aorta, left external carotid artery, right external carotid artery, left subclavian artery, right subclavian artery, left internal carotid artery, right internal carotid artery, left vertebral artery, right vertebral artery. On this basis, the left internal carotid artery, right internal carotid artery, left vertebral artery, and right vertebral artery can be default-specified as the target categories. Of course, the user can also specify at least one of them as the target category. For example, the user specifies the left internal carotid artery, right internal carotid artery, left vertebral artery, and right vertebral artery as the target categories. Other cases can be inferred by analogy and will not be exemplified one by one here.

[0065] In one implementation scenario, specifically, based on the vascular paths of each vascular category, the vascular regions that do not belong to the target category can be cropped to obtain a second image region. Based on this, for each pixel point in the second image region, the pixel points that meet the preset conditions can be removed based on the deviation between the first distance corresponding to the pixel point and the second distance corresponding to the pixel point, so as to obtain a target vascular region belonging to the target category; where the first distance represents the shortest distance from the pixel point to each first pixel point, the second distance represents the shortest distance from the pixel point to the target path, and the target path represents the vascular path of the target category. It should be noted that before cropping, the first distance can be measured first, and after cropping, the second distance can be measured again. Or, before cropping, the first distance and the second distance can be measured first, which is not limited here. The above method can, on the one hand, cut off the vascular regions of irrelevant categories, and on the other hand, by measuring the deviation between the first distance and the second distance and removing the pixel points that meet the preset conditions, the vascular region can be further optimized, which is beneficial to improving the accuracy of the vascular region.

[0066] In a specific implementation scenario, still taking the head and neck region as an example, as mentioned above, the several preset categories can include, but are not limited to: aorta, left external carotid artery, right external carotid artery, left subclavian artery, right subclavian artery, left internal carotid artery, right internal carotid artery, left vertebral artery, right vertebral artery, and the target categories can include, but are not limited to: left internal carotid artery, right internal carotid artery, left vertebral artery, right vertebral artery. Then the irrelevant categories can include, but are not limited to: aorta, left external carotid artery, right external carotid artery, left subclavian artery, right subclavian artery, etc., which is not limited here. In addition, other cases can be inferred by analogy and will not be exemplified one by one here.

[0067] In a specific implementation scenario, when the blood vessel path that does not belong to the target category does not intersect with the blood vessel path of the target category, the pixel points in the first image region that do not belong to the target category can be directly excluded from the first image region to complete the cropping of the blood vessel region that does not belong to the target category; while in the case where the blood vessel path that does not belong to the target category intersects with the blood vessel path of the target category, the included angle between the blood vessel path that does not belong to the target category and the blood vessel path of the target category at the intersection can be combined, and the pixel points that do not belong to the target category can be excluded at the intersection, and the remaining region can be appropriately repaired after the exclusion. The specific process will not be elaborated here.

[0068] In a specific implementation scenario, for each pixel point in the second image region, the shortest distances from the pixel point to each first pixel point and to the target path can be measured by means such as Euclidean distance transformation. For the specific calculation method, the technical details of distance measurement methods such as Euclidean distance transformation can be referred to, which will not be elaborated here.

[0069] In a specific implementation scenario, the distance measurement for each pixel point in the second image region can be calculated in parallel by a GPU (Graphic Processor Unit, image processing unit). That is to say, the deviation corresponding to each pixel point in the second image region can be calculated in parallel by the GPU, which is beneficial to greatly improving the speed of distance measurement.

[0070] In a specific implementation scenario, the blood vessel path is represented by the blood vessel centerline. To further improve the accuracy of the blood vessel region of the target category, before measuring the second distance, each first pixel point on the target path can be used as the current pixel point, and the truncation reference of the normal reference of the target path at the current pixel point in the second image region can be obtained. It should be noted that when the blood vessel image to be measured is a three-dimensional image, the normal reference is the normal plane and the truncation reference is the truncation plane; while when the blood vessel image to be measured is a two-dimensional image, the normal reference is the normal line and the truncation reference is the truncation line. On this basis, the pixel point located at the center position of the truncation reference is used as the second pixel point corresponding to the current pixel point, and the second pixel points corresponding to each first pixel point are connected to obtain an optimized target path. For example, the second pixel points corresponding to each first pixel point can be connected in the connection order of each first pixel point to obtain an optimized target path. The above method can improve the accuracy of the target path and is beneficial to the accuracy of the finally optimized blood vessel region.

[0071] In a specific implementation scenario, the above deviation can specifically be the absolute value of the difference between the first distance and the second distance. In this case, the preset condition can be set as the deviation being greater than the distance threshold, and the distance threshold can be set according to the actual situation and is not limited here. That is to say, pixel points with a relatively large deviation can be excluded from the target second image area. It should be noted that when cropping the blood vessel area in the first image area that does not belong to the target category, it may not be cropped clean due to interference in the first image area (for example, there may be pixel points in the second image area that are misidentified as the target category but actually do not belong to the target category). Since for pixel points that truly belong to the target category, the deviation between the corresponding first distance and the second distance should be relatively small, secondary cropping through the above deviation at this time can help improve the accuracy of the target blood vessel area of the target category.

[0072] In an implementation scenario, after obtaining the target blood vessel area belonging to the target category, a curved surface reconstruction can also be performed based on the target blood vessel area of the target category to obtain a CPR (Curved-Planar Reconstruction) image sequence of the target category. Still taking the head and neck area as an example, after obtaining the blood vessel areas of the internal carotid artery and the vertebral artery, a CPR image sequence of the internal carotid artery and the vertebral artery can be reconstructed. For the specific reconstruction process, the technical details of CPR can be referred to and will not be elaborated here.

[0073] For the above solution, contour detection is performed on the blood vessel image to be measured to obtain the first image area of the blood vessel, and the blood vessel category to which each pixel point in the first image area belongs is identified. Additionally, several first pixel points representing the blood vessel path are extracted from the first image area, and the blood vessel category includes several preset categories. On this basis, based on the reference pixel points belonging to the reference category, the paths of the first pixel points belonging to the same blood vessel category are analyzed respectively to obtain the blood vessel paths of each blood vessel category. The blood vessel of the reference category is in a preset position among the blood vessels of several preset categories. And based on the blood vessel path, the first image area is optimized to obtain the target blood vessel area belonging to the target category. On the one hand, since there is no need to manually process and analyze the image to obtain the blood vessel area, it helps to improve the image detection efficiency. On the other hand, during the path analysis process, by using the reference pixel points subordinate to the blood vessel of the reference category that is in a preset position among the blood vessels of several preset categories to analyze the first pixel points belonging to the same blood vessel category to obtain the blood vessel path of this blood vessel category, it helps to exclude interference as much as possible and improve the accuracy of image detection.

[0074] Please refer to Figure 3 , Figure 3It is a framework schematic diagram of an embodiment of the image processing apparatus 30 of the present application. The image processing apparatus 30 includes: a contour detection module 31, a category recognition module 32, a path analysis module 33, and a region optimization module 34. The contour detection module 31 is configured to perform contour detection on the blood vessel image to be measured to obtain a first image region of the blood vessel. The category recognition module 32 is configured to identify the blood vessel category to which each pixel point in the first image region belongs, and extract a plurality of first pixel points representing the blood vessel path in the first image region. Among them, the blood vessel category includes a plurality of preset categories. The path analysis module 33 is configured to perform path analysis on the first pixel points belonging to the same blood vessel category respectively based on the reference pixel points belonging to the reference category, to obtain the blood vessel paths of each blood vessel category. Among them, the blood vessel of the reference category is in a preset position among the blood vessels of the plurality of preset categories. The region optimization module 34 is configured to optimize the first image region based on the blood vessel path to obtain a target blood vessel region belonging to the target category.

[0075] In the above solution, on the one hand, since there is no need to manually process and analyze the image to obtain the blood vessel region, it is beneficial to improve the image detection efficiency. On the other hand, in the path analysis process, the reference pixel points subordinate to the blood vessels of the reference category in a preset position among the blood vessels of the plurality of preset categories are used to analyze the first pixel points belonging to the same blood vessel category to obtain the blood vessel path of this blood vessel category, which is beneficial to exclude interference as much as possible and improve the accuracy of image detection.

[0076] In some disclosed embodiments, the path analysis module 33 includes a category selection sub-module for respectively taking each blood vessel category as the current category. The path analysis module 33 includes an end point determination sub-module for determining the end point of the blood vessel path of the current category based on the distances from each first pixel point belonging to the current category to the reference pixel points. The path analysis module 33 includes a path acquisition sub-module for sequentially connecting each first pixel point belonging to the current category based on the end point of the blood vessel path to obtain the blood vessel path of the current category.

[0077] Therefore, during the path analysis process, the end point of the blood vessel path can be determined first, and then the blood vessel path can be obtained by connecting each first pixel point according to the end point of the blood vessel path, which can avoid the influence of other categories on the path analysis of the current category as much as possible and is beneficial to improving the detection speed.

[0078] In some disclosed embodiments, the preset position is the central position, and the end point of the blood vessel path is the first pixel point with the maximum distance; and / or, each first pixel point belonging to the current category is sequentially connected by using the shortest path algorithm.

[0079] Therefore, setting the preset position to the center position and setting the end point of the blood vessel path to the first pixel point with the maximum distance can greatly reduce the complexity of determining the end point of the blood vessel path and is beneficial to further improving the detection speed; and using the shortest path algorithm to sequentially connect each first pixel point belonging to the current category to obtain the blood vessel path can greatly reduce the probability of errors in connecting the first pixel points and improve the accuracy of the blood vessel path.

[0080] In some disclosed embodiments, the path analysis module 33 includes a path checking sub-module for checking whether there is a blood vessel path of the same blood vessel category; the path analysis module 33 includes a path connecting sub-module for connecting the blood vessel paths of the same blood vessel category in response to the existence of a blood vessel path of the same blood vessel category.

[0081] Therefore, after obtaining the blood vessel path of the current category, further checking whether there is a blood vessel path of the same blood vessel category and connecting the blood vessel paths of the same blood vessel category in response to the existence of a blood vessel path of the same blood vessel category can greatly alleviate the problem of blood vessel image breakage caused by reasons such as blood vessel stenosis and angiography, which is beneficial to improving the detection accuracy.

[0082] In some disclosed embodiments, the region optimization module 34 includes a region cropping sub-module for cropping the blood vessel regions that do not belong to the target category based on the blood vessel paths of each blood vessel category to obtain a second image region; the region optimization module 34 includes a pixel rejection sub-module for rejecting the pixel points whose deviations meet the preset conditions for each pixel point in the second image region based on the deviation between the first distance corresponding to the pixel point and the second distance corresponding to the pixel point, to obtain a target blood vessel region belonging to the target category; wherein, the first distance represents the shortest distance from the pixel point to each first pixel point, the second distance represents the shortest distance from the pixel point to the target path, and the target path represents the blood vessel path of the target category.

[0083] Therefore, on the one hand, it can cut off the blood vessel regions of irrelevant categories, and on the other hand, by measuring the deviation between the first distance and the second distance and rejecting the pixel points whose deviations meet the preset conditions, the blood vessel region can be further optimized, which is beneficial to improving the accuracy of the blood vessel region.

[0084] In some disclosed embodiments, the blood vessel path is represented by the blood vessel center line. The region optimization module 34 includes a pixel selection sub-module for respectively taking each first pixel point on the target path as the current pixel point; the region optimization module 34 includes a region truncation sub-module for obtaining the truncation reference of the normal reference of the target path at the current pixel point in the second image region; the region optimization module 34 includes a center determination sub-module for taking the pixel point located at the center position of the truncation reference as the second pixel point corresponding to the current pixel point; the region optimization module 34 includes a path optimization sub-module for connecting the second pixel points corresponding to each first pixel point to obtain the optimized target path; wherein, when the blood vessel image to be measured is a three-dimensional image, the normal reference is a normal plane and the truncation reference is a truncation plane, and when the blood vessel image to be measured is a two-dimensional image, the normal reference is a normal line and the truncation reference is a truncation line.

[0085] Therefore, further optimizing the target path before measuring the second distance can improve the accuracy of the target path, which is beneficial to the accuracy of the finally optimized blood vessel region.

[0086] In some disclosed embodiments, the deviation is the absolute value of the difference between the first distance and the second distance, and the preset condition includes that the deviation is greater than the distance threshold; and / or, the deviation corresponding to the pixel point is calculated in parallel by the GPU.

[0087] Therefore, the deviation is set as the absolute value of the difference between the first distance and the second distance, and the preset condition is set to include that the deviation is greater than the distance threshold, so that pixel points with too large deviations can be eliminated, which is beneficial to reducing the complexity of eliminating pixel points. Calculating the deviation corresponding to each pixel point in parallel by the GPU can improve the overall calculation speed of the deviation, thus being beneficial to improving the detection efficiency.

[0088] In some disclosed embodiments, the category recognition module 32 includes a data acquisition sub-module for obtaining first image data based on the pixel values of each pixel point in the blood vessel image to be measured respectively, and obtaining second image data based on the offset values of each pixel point from the reference position respectively; wherein, the first image data includes the pixel values of each pixel point, and the second image data includes the offset values corresponding to each pixel point; the category recognition module 32 includes a data fusion sub-module for fusing the first image data and the second image data to obtain fused image data; the category recognition module 32 includes a category recognition sub-module for using the recognition model to recognize the fused image data to obtain the blood vessel category to which each pixel point belongs.

[0089] Therefore, in the process of category recognition, on the one hand, the texture feature information represented by the first image data is referred to, and on the other hand, the geometric feature information represented by the second image data can be referred to, which is beneficial to improving the accuracy of blood vessel category recognition.

[0090] In some disclosed embodiments, the category recognition module 32 includes a sub-image extraction sub-module for extracting a plurality of sub-image data from the fused image data; wherein, the set of the plurality of sub-image data covers the first image region; the category recognition sub-module includes a sub-image recognition unit for respectively recognizing the plurality of sub-image data by using a recognition model to obtain the category recognition results corresponding to the respective sub-image data; wherein, the category recognition results include the probability values of the pixel points belonging to a plurality of preset categories respectively; the category recognition sub-module includes a category determination unit for obtaining the blood vessel categories to which the respective pixel points belong based on the category recognition results corresponding to the respective sub-image data.

[0091] Therefore, by converting the overall recognition of the fused image data into the sub-image recognition of the respective sub-image data, it is beneficial to greatly reduce the computing load and improve the detection speed.

[0092] In some disclosed embodiments, the plurality of preset categories include: the aorta, the left external carotid artery, the right external carotid artery, the left subclavian artery, the right subclavian artery, the left internal carotid artery, the right internal carotid artery, the left vertebral artery, and the right vertebral artery. The reference category includes the aorta, and the target categories include: the left internal carotid artery, the right internal carotid artery, the left vertebral artery, and the right vertebral artery.

[0093] Therefore, by setting the plurality of preset categories to include: the aorta, the left external carotid artery, the right external carotid artery, the left subclavian artery, the right subclavian artery, the left internal carotid artery, the right internal carotid artery, the left vertebral artery, and the right vertebral artery, and setting the reference category to include the aorta, and setting the target categories to include: the left internal carotid artery, the right internal carotid artery, the left vertebral artery, and the right vertebral artery, it can be applied to the blood vessel detection process of the head and neck arteries, and detect and obtain the blood vessel regions of the internal carotid artery and the vertebral artery, which is beneficial to providing rich auxiliary information in the head and neck region.

[0094] Please refer to Figure 4 , Figure 4 which is a schematic framework diagram of an embodiment of the electronic device 40 of the present application. The electronic device 40 includes a memory 41 and a processor 42 that are coupled to each other. The processor 42 is configured to execute program instructions stored in the memory 41 to implement the steps of any of the above embodiments of the image processing method. In a specific implementation scenario, the electronic device 40 may include, but is not limited to: a microcomputer, a server. In addition, the electronic device 40 may also include mobile devices such as a laptop computer and a tablet computer, which are not limited herein.

[0095] Specifically, the processor 42 is used to control itself and the memory 41 to implement the steps of any of the above embodiments of the image processing method. The processor 42 may also be referred to as a CPU (Central Processing Unit). The processor 42 may be an integrated circuit chip with signal processing capabilities. The processor 42 may also be a general-purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. Additionally, the processor 42 may be implemented jointly by integrated circuit chips.

[0096] In the above solution, on the one hand, since there is no need to manually process and analyze the image to obtain the blood vessel region, it is beneficial to improve the image detection efficiency. On the other hand, in the path analysis process, by using the reference pixel points subordinate to the blood vessels of the reference category at the preset position among several preset categories of blood vessels, the blood vessel path of the first pixel points belonging to the same blood vessel category is analyzed, which is beneficial to excluding interference as much as possible and improving the accuracy of image detection.

[0097] Please refer to Figure 5 , Figure 5 , which is a schematic framework diagram of an embodiment of the computer-readable storage medium 50 of the present application. The computer-readable storage medium 50 stores program instructions 501 that can be run by a processor, and the program instructions 501 are used to implement the steps of any of the above embodiments of the image processing method.

[0098] In the above solution, on the one hand, since there is no need to manually process and analyze the image to obtain the blood vessel region, it is beneficial to improve the image detection efficiency. On the other hand, in the path analysis process, by using the reference pixel points subordinate to the blood vessels of the reference category at the preset position among several preset categories of blood vessels, the blood vessel path of the first pixel points belonging to the same blood vessel category is analyzed, which is beneficial to excluding interference as much as possible and improving the accuracy of image detection.

[0099] In several embodiments provided by the present application, it should be understood that the disclosed methods and apparatuses can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some interfaces. The indirect couplings or communication connections of the apparatuses or units can be in electrical, mechanical or other forms.

[0100] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0101] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0102] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods in each embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0103] If the technical solution of this application involves personal information, the product using the technical solution of this application has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using the technical solution of this application has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, the personal information processing rules are notified by obvious signs / information, and the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. An image processing method, characterized in that, Including: Performing contour detection on the blood vessel image to be measured to obtain a first image region of the blood vessel; Identifying the blood vessel category to which each pixel point in the first image region belongs, and extracting a plurality of first pixel points representing the blood vessel path in the first image region; wherein, the blood vessel category includes a plurality of preset categories; Based on the reference pixel points belonging to the reference category, respectively performing path analysis on the first pixel points belonging to the same blood vessel category to obtain the blood vessel paths of each blood vessel category; wherein, the blood vessels of the reference category are in a preset position among the blood vessels of the plurality of preset categories; Based on the blood vessel path, optimizing the first image region to obtain a target blood vessel region belonging to the target category, including: based on the blood vessel paths of each blood vessel category, cropping the blood vessel regions that do not belong to the target category to obtain a second image region; for each pixel point in the second image region, based on the deviation between the first distance corresponding to the pixel point and the second distance corresponding to the pixel point, removing the pixel points whose deviation meets the preset conditions to obtain a target blood vessel region belonging to the target category; wherein, the first distance represents the shortest distance from the pixel point to each of the first pixel points, the second distance represents the shortest distance from the pixel point to the target path, and the target path represents the blood vessel path of the target category.

2. The method according to claim 1, characterized in that, The step of respectively performing path analysis on the first pixel points belonging to the same blood vessel category based on the reference pixel points belonging to the reference category to obtain the blood vessel paths of each blood vessel category includes: Taking each blood vessel category as the current category respectively; Based on the distances from each of the first pixel points belonging to the current category to the reference pixel points, determining the end point of the blood vessel path of the current category; Based on the end point of the blood vessel path, sequentially connecting each of the first pixel points belonging to the current category to obtain the blood vessel path of the current category.

3. The method according to claim 2, wherein The preset position is the central position, and the end point of the blood vessel path is the first pixel point with the maximum distance; And / or, each of the first pixel points belonging to the current category is sequentially connected by using the shortest path algorithm.

4. The method according to claim 2, wherein After the step of based on the end point of the blood vessel path, sequentially connecting each of the first pixel points belonging to the current category to obtain the blood vessel path of the current category, the method further includes: Checking whether there is a blood vessel path of the same blood vessel category; In response to the existence of a blood vessel path of the same blood vessel category, connecting the blood vessel paths of the same blood vessel category.

5. The method according to claim 1, characterized in that, characterized in that, The blood vessel path is represented by the blood vessel center line, and the method includes: Respectively taking each of the first pixel points on the target path as the current pixel point; Obtaining the truncation reference of the normal reference of the target path at the current pixel point in the second image region; Taking the pixel point located at the central position of the truncation reference as the second pixel point corresponding to the current pixel point; Connecting the second pixel points corresponding to each of the first pixel points to obtain the optimized target path; Wherein, when the blood vessel image to be measured is a three-dimensional image, the normal reference is a normal plane, and the truncation reference is a truncation plane; when the blood vessel image to be measured is a two-dimensional image, the normal reference is a normal line, and the truncation reference is a truncation line.

6. The method according to claim 1, characterized in that, The deviation is the absolute value of the difference between the first distance and the second distance, and the preset condition includes that the deviation is greater than a distance threshold; and / or, the deviation corresponding to the pixel point is calculated in parallel by a GPU.

7. The method according to claim 1, wherein The identifying the blood vessel category to which each pixel point in the first image region belongs includes: obtaining first image data based on the pixel values of each of the pixel points in the blood vessel image to be measured, and obtaining second image data based on the offset values of each of the pixel points from a reference position; wherein, the first image data includes the pixel values of each of the pixel points, and the second image data includes the offset values corresponding to each of the pixel points; fusing the first image data and the second image data to obtain fused image data; using an identification model to identify the fused image data to obtain the blood vessel category to which each of the pixel points belongs.

8. The method according to claim 7, wherein Before the using the identification model to identify the fused image data to obtain the blood vessel category to which each of the pixel points belongs, the method further includes: extracting a plurality of sub-image data from the fused image data; wherein, the set of the plurality of sub-image data covers the first image region; The using the identification model to identify the fused image data to obtain the blood vessel category to which each of the pixel points belongs includes: using the identification model to respectively identify the plurality of sub-image data to obtain category identification results corresponding to each of the sub-image data; wherein, the category identification results include the probability values of the pixel points belonging to the plurality of preset categories respectively; obtaining the blood vessel category to which each of the pixel points belongs based on the category identification results corresponding to each of the sub-image data.

9. The method according to any one of claims 1 to 8, characterized in that, The plurality of preset categories include: aorta, left external carotid artery, right external carotid artery, left subclavian artery, right subclavian artery, left internal carotid artery, right internal carotid artery, left vertebral artery, right vertebral artery, the reference category includes the aorta, and the target categories include: left internal carotid artery, right internal carotid artery, left vertebral artery, right vertebral artery.

10. An image processing apparatus, characterized in that, including: a contour detection module, configured to perform contour detection on a blood vessel image to be measured to obtain a first image region of the blood vessel; a category identification module, configured to identify the blood vessel category to which each pixel point in the first image region belongs, and extract a plurality of first pixel points representing a blood vessel path in the first image region; wherein, the blood vessel category includes a plurality of preset categories; a path analysis module, configured to perform path analysis on the first pixel points belonging to the same blood vessel category respectively based on reference pixel points belonging to a reference category to obtain blood vessel paths of each of the blood vessel categories; wherein, the blood vessel of the reference category is in a preset position among the blood vessels of the plurality of preset categories; A region optimization module, configured to optimize the first image region based on the blood vessel path to obtain a target blood vessel region belonging to a target category, including: cropping the blood vessel regions that do not belong to the target category based on the blood vessel paths of each blood vessel category to obtain a second image region; for each pixel point in the second image region, removing the pixel points whose deviation meets a preset condition based on the deviation between the first distance corresponding to the pixel point and the second distance corresponding to the pixel point, to obtain a target blood vessel region belonging to the target category; wherein, the first distance represents the closest distance from the pixel point to each of the first pixel points, and the second distance represents the closest distance from the pixel point to the target path, and the target path represents the blood vessel path of the target category.

11. An electronic device, characterized in that, It includes a memory and a processor coupled to each other, and the processor is configured to execute program instructions stored in the memory to implement the image processing method according to any one of claims 1 to 9.

12. A computer-readable storage medium having program instructions stored thereon, characterized in that, When the program instructions are executed by the processor, the image processing method according to any one of claims 1 to 9 is implemented.

Images