Image-based neurology lesion auxiliary identification method
Through multi-angle image processing and neural network technology, the problems of low efficiency and poor accuracy in vascular stenosis identification are solved, and efficient and accurate vascular stenosis identification is achieved.
Patent Information
- Application Number
- CN202510915624.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-03
- Publication Date
- 2025-10-03
AI Technical Summary
In the prior art, doctors rely on experience to identify vascular stenosis in angiography images, which is inefficient and inaccurate, and manual identification is prone to errors.
A cross-Transformer denoising convolutional neural network was used to denoise angiography images. The Laplace and Sobel operators were combined for vessel segmentation and edge extraction. Multi-angle image data were used to correct vessel width, and potential abnormal areas were identified by analyzing the width change trend.
The accuracy and efficiency of vascular stenosis identification are improved, human errors are reduced, data correction corrects the impact of vascular deformation, and improves the accuracy of image recognition.
Smart Images
Figure CN120748017A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data management, and in particular to an image-based auxiliary identification method for neurological lesions. Background Art
[0002] Neurology is a secondary discipline specializing in neurological diseases, primarily treating cerebrovascular diseases, migraines, inflammatory brain diseases, myelitis, epilepsy, dementia, metabolic diseases, genetic predisposition diseases, trigeminal neuralgia, sciatica, peripheral neuropathy, and myasthenia gravis. Cerebrovascular diseases include cerebral thrombosis and cerebral embolism, both of which can reduce or obstruct blood flow, causing vascular stenosis.
[0003] In the existing technology, doctors observe angiography images and rely on their own experience to identify vascular stenosis. However, this process is too inefficient and overly dependent on the doctor's experience. In addition, the blood vessels in angiography images are densely distributed, making manual identification prone to errors. Summary of the Invention
[0004] In order to solve the technical problems of low efficiency and poor accuracy in manual identification, the image-based auxiliary identification method for neurological lesions proposed in the present invention specifically includes the following steps:
[0005] S1. Acquire a first angiographic image, a second angiographic image, and a third angiographic image at different shooting angles, wherein the shooting angle of the first angiographic image is between the shooting angles of the second angiographic image and the third angiographic image, and the shooting angles of the second angiographic image and the third angiographic image are symmetrically distributed with respect to the shooting angle of the first angiographic image;
[0006] S2. performing blood vessel segmentation on the first angiography image;
[0007] S3. Determine blood vessel width data in the first angiography image;
[0008] S4. Determine blood vessel width data in the second angiography image based on the blood vessel segmentation result of the first angiography image;
[0009] S5. Determine blood vessel width data in the third angiography image based on the blood vessel segmentation result of the first angiography image;
[0010] S6. Correcting the blood vessel width data in the first angiographic image based on the blood vessel width data in the second angiographic image and the blood vessel width data in the third angiographic image;
[0011] S7. Determine a vascular abnormality region in the first angiography image based on the vascular width data.
[0012] Preferably, in S2, a cross Transformer denoising convolutional neural network is used to denoise the first angiography image, a segmentation threshold is determined, the first angiography image is binarized according to the segmentation threshold to obtain an edge extraction image, edge points are extracted from the edge extraction image, and the blood vessel contour is extracted based on the edge points.
[0013] Preferably, in S2, the segmentation threshold is determined by using a Laplace operator to extract edge points from the first angiography image according to image gradient information, calculating the grayscale mean of all edge points, and taking half of the grayscale mean as the segmentation threshold.
[0014] Preferably, in S2, the specific process of extracting edge points from the edge extraction image is to use a Sobel operator to extract edge points from the edge extraction image according to image gradient information.
[0015] Preferably, in S2, the specific process of extracting the blood vessel contour based on edge points is that the doctor marks the basic shape of the blood vessel in the edge extraction image, connects the edge points under the guidance of the basic shape, determines the blood vessel contour, and uses the trunk of the blood vessel contour and its various branches as the various first blood vessel areas.
[0016] Preferably, in S3, the first blood vessel region is divided into a plurality of first blood vessel sub-regions according to the extending direction of the blood vessel, and the region width of each first blood vessel sub-region is determined, so that each first blood vessel region corresponds to a set of region width data.
[0017] Preferably, in S4, a cross-Transformer denoising convolutional neural network is used to perform denoising processing on the second angiography image, and a Sobel operator is used to extract edge points from the denoised second angiography image based on image gradient information. Edge points whose distance from the blood vessel contour of the first angiography image is less than a distance threshold are regarded as valid edge points, and the valid edge points are connected to determine the blood vessel contour of the second angiography image. The trunk and each branch of the blood vessel contour of the second angiography image are regarded as each second blood vessel region. Based on the first blood vessel region, the second blood vessel region is divided into multiple second blood vessel sub-regions according to the extension direction of the blood vessel, and the region width of each second blood vessel sub-region is determined, so that each second blood vessel region corresponds to a set of region width data.
[0018] Preferably, in S5, a cross-Transformer denoising convolutional neural network is used to perform denoising processing on the third angiography image, and a Sobel operator is used to extract edge points from the denoised third angiography image based on image gradient information. Edge points whose distance from the blood vessel contour of the first angiography image is less than a distance threshold are regarded as valid edge points, and the valid edge points are connected to determine the blood vessel contour of the third angiography image. The trunk and each branch of the blood vessel contour of the third angiography image are regarded as each third blood vessel region. The third blood vessel region is divided into a plurality of third blood vessel sub-regions according to the extension direction of the blood vessel based on the first blood vessel region, and the region width of each third blood vessel sub-region is determined, so that each third blood vessel region corresponds to a set of region width data.
[0019] Preferably, in S6, for each first blood vessel region, the corresponding second blood vessel region and third blood vessel region are found, and the region width data of the first blood vessel region is updated according to the corresponding second blood vessel region and third blood vessel region until the region width data of all first blood vessel regions are updated.
[0020] Preferably, in S7, the change trend of the regional width data corresponding to the first blood vessel region is analyzed. If there is a width reduction trend or a width increase trend in the change trend, the first blood vessel region is marked as a potential abnormal region; otherwise, the first blood vessel region is marked as a normal region.
[0021] Compared with the prior art, the present invention has the following beneficial effects:
[0022] Shoot at both sides of the main shooting angle respectively, use the data from both sides to correct the data of the main shooting angle, correct the blood vessel deformation, and make the data more accurate. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1 It is a flow chart of the neurological lesion auxiliary identification method of the present invention. DETAILED DESCRIPTION
[0024] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and specific implementation methods.
[0025] like Figure 1 As shown, the image-based neurological lesion auxiliary identification method proposed by the present invention specifically includes the following steps:
[0026] S1. Acquire a first angiographic image, a second angiographic image, and a third angiographic image at different shooting angles. The shooting angle of the first angiographic image is between the shooting angles of the second and third angiographic images, and the shooting angles of the second and third angiographic images are symmetrically distributed relative to the shooting angle of the first angiographic image. Specifically, the first, second, and third angiographic images are acquired using CT vascular enhancement technology. The shooting angle of the second angiographic image is 10° to the left of the shooting angle of the first angiographic image, and the shooting angle of the third angiographic image is 10° to the right of the shooting angle of the first angiographic image.
[0027] S2. Perform blood vessel segmentation on the first angiography image. Specifically, a cross-Transformer denoising convolutional neural network is used to denoise the first angiography image, determine a segmentation threshold, binarize the first angiography image according to the segmentation threshold, obtain an edge extraction image, extract edge points from the edge extraction image, and extract the blood vessel contour based on the edge points. The process of determining the segmentation threshold is to use the Laplace operator to extract edge points from the first angiography image according to the image gradient information, calculate the grayscale mean of all edge points, and use half of the grayscale mean as the segmentation threshold. The specific process of extracting edge points from the edge extraction image is to use the Sobel operator to extract edge points from the edge extraction image according to the image gradient information. The specific process of extracting the blood vessel contour based on edge points is that the blood vessels often have a tree-like structure. The doctor marks the basic shape of the blood vessel in the edge extraction image, connects the edge points under the guidance of the basic shape, and determines the blood vessel contour. Since the blood vessel contour is a tree-like structure, the trunk of the blood vessel contour and its various branches are regarded as the various first blood vessel regions.
[0028] S3. Determine vessel width data in the first angiography image. Specifically, divide the first vessel region into multiple first vessel sub-regions according to the vessel extension direction, and determine the region width of each first vessel sub-region, so that each first vessel region corresponds to a set of region width data. The first vessel sub-region division process is as follows: the first vessel region in the image includes two vessel walls extending along the vessel extension direction. The vessel walls are divided into a set number of equally spaced partitions, and the corresponding partition points in the two vessel walls are connected to determine each first vessel sub-region. The region width of the first vessel sub-region is calculated by obtaining the region start point and region end point of each first vessel sub-region on each vessel wall, calculating the vertical distance between the region start points and the vertical distance between the region end points, and taking the average of the two values as the region width of the first vessel sub-region.
[0029] S4. Determine vessel width data in the second angiographic image based on the vessel segmentation result of the first angiographic image. Specifically, a cross-Transformer denoising convolutional neural network is used to denoise the second angiographic image. Edge points are extracted from the denoised second angiographic image using the Sobel operator based on image gradient information. Edge points whose distance from the vessel contour of the first angiographic image is less than a threshold are considered valid edge points. The valid edge points are connected to determine the vessel contour of the second angiographic image. The trunk and its branches of the vessel contour of the second angiographic image are defined as second vessel regions. Based on the first vessel region, the second vessel region is divided into a plurality of second vessel sub-regions according to the direction of the vessel extension. The width of each second vessel sub-region is determined, such that each second vessel region corresponds to a set of region width data. The second vessel sub-region division process includes determining the first vessel region corresponding to the second vessel region, using the number of first vessel sub-regions in the first vessel region as the preset number of partitions of the current second vessel region, dividing the vessel wall into equal intervals according to the preset number of partitions, and connecting the corresponding partition points in the two vessel walls to determine each second vessel sub-region. The calculation process of the area width of the second vessel sub-area is the same as that of the first vessel sub-area, and will not be repeated here.
[0030] S5. Determine vessel width data in the third angiographic image based on the vessel segmentation result of the first angiographic image. Specifically, a cross-Transformer denoising convolutional neural network is used to denoise the third angiographic image. Edge points are extracted from the denoised third angiographic image using the Sobel operator based on image gradient information. Edge points whose distance from the vessel contour of the first angiographic image is less than a threshold are considered valid edge points. The valid edge points are connected to determine the vessel contour of the third angiographic image. The trunk and its branches of the vessel contour of the third angiographic image are defined as third vessel regions. Based on the first vessel region, the third vessel region is divided into a plurality of third vessel sub-regions according to the direction of the vessel extension. The width of each third vessel sub-region is determined, such that each third vessel region corresponds to a set of region width data. The third vessel sub-region division process includes determining the first vessel region corresponding to the third vessel region, using the number of first vessel sub-regions in the first vessel region as the preset number of partitions of the current third vessel region, dividing the vessel wall into equal intervals according to the preset number of partitions, and connecting the corresponding partition points in two vessel walls to determine each third vessel sub-region. The calculation process of the region width of the third vessel sub-region is the same as that of the first vessel sub-region, and will not be repeated here.
[0031] S6. Correct the vessel width data in the first angiographic image based on the vessel width data in the second and third angiographic images. Specifically, for each first vessel region, find the corresponding second and third vessel regions, and update the region width data of the first vessel region based on the corresponding second and third vessel regions, until the region width data of all first vessel regions is updated. The region width data updating process for the first vessel region involves, for each first vessel sub-region, finding the corresponding second and third vessel sub-regions within the corresponding second and third vessel regions of the first vessel region, calculating the average of the region widths of the three sub-regions, and using this average as the updated region width of the first vessel sub-region, until the region width data of all first vessel sub-regions is updated.
[0032] S7. Determine an abnormal vascular region in the first angiographic image based on the vascular width data. Specifically, analyze the changing trend of the region width data corresponding to the first vascular region. If the changing trend shows a decreasing or increasing width trend, it indicates that the first vascular region may have vascular stenosis, and the first vascular region is marked as a potential abnormal region. Otherwise, the first vascular region is marked as a normal region. The changing trend is determined by calculating the difference between the previous and next region widths. If the difference is negative and the absolute value of the difference is greater than a judgment threshold, the width is increasing. If the difference is positive and the absolute value of the difference is greater than the judgment threshold, the width is decreasing.
[0033] Due to the limitation of shooting angle, angiography images may cause a certain degree of deformation of the photographed blood vessels, affecting the accuracy. The present invention shoots at two main shooting angles respectively, and uses the data from both sides to correct the data of the main shooting angle, correct the blood vessel deformation, and make the data more accurate.
[0034] The above disclosure is only a preferred embodiment of the present invention and is not intended to limit the scope of the present invention. It should be noted that for those skilled in the art, any equivalent changes made to the present invention without departing from the design structure and principles of the present invention are considered to be within the scope of protection of the present invention.
Claims
1. An image-based auxiliary identification method for neurological lesions, characterized in that: The neurological lesion auxiliary identification method specifically comprises the following steps: S1. Acquire a first angiographic image, a second angiographic image, and a third angiographic image at different shooting angles, wherein the shooting angle of the first angiographic image is between the shooting angles of the second angiographic image and the third angiographic image, and the shooting angles of the second angiographic image and the third angiographic image are symmetrically distributed with respect to the shooting angle of the first angiographic image; S2. performing blood vessel segmentation on the first angiography image; S3. Determine blood vessel width data in the first angiography image; S4. Determine blood vessel width data in the second angiography image based on the blood vessel segmentation result of the first angiography image; S5. Determine blood vessel width data in the third angiography image based on the blood vessel segmentation result of the first angiography image; S6. Correcting the blood vessel width data in the first angiographic image based on the blood vessel width data in the second angiographic image and the blood vessel width data in the third angiographic image; S7. Determine a vascular abnormality region in the first angiography image based on the vascular width data.
2. The neurological lesion auxiliary identification method according to claim 1, characterized in that: In S2, a cross-Transformer denoising convolutional neural network is used to denoise the first angiography image, a segmentation threshold is determined, the first angiography image is binarized according to the segmentation threshold to obtain an edge extraction image, edge points are extracted from the edge extraction image, and blood vessel contours are extracted based on the edge points.
3. The neurological lesion auxiliary identification method according to claim 2, characterized in that: In S2, the segmentation threshold is determined by using a Laplace operator to extract edge points from the first angiography image according to image gradient information, calculating the grayscale mean of all edge points, and using half of the grayscale mean as the segmentation threshold.
4. The neurological lesion auxiliary identification method according to claim 2, characterized in that: In S2, the specific process of extracting edge points from the edge extraction image is to use the Sobel operator to extract edge points from the edge extraction image according to image gradient information.
5. The neurological lesion auxiliary identification method according to claim 2, characterized in that: In S2, the specific process of extracting the blood vessel contour based on edge points is that the doctor marks the basic shape of the blood vessel in the edge extraction image, connects the edge points under the guidance of the basic shape, determines the blood vessel contour, and uses the trunk of the blood vessel contour and its various branches as the various first blood vessel regions.
6. The neurological lesion auxiliary identification method according to claim 5, characterized in that: In S3, the first blood vessel region is divided into a plurality of first blood vessel sub-regions according to the extending direction of the blood vessel, and the region width of each first blood vessel sub-region is determined, so that each first blood vessel region corresponds to a set of region width data.
7. The neurological lesion auxiliary identification method according to claim 6, characterized in that: In S4, a cross-Transformer denoising convolutional neural network is used to denoise the second angiography image. A Sobel operator is used to extract edge points from the denoised second angiography image based on image gradient information. Edge points whose distance from the blood vessel contour of the first angiography image is less than a distance threshold are defined as valid edge points. The valid edge points are connected to determine the blood vessel contour of the second angiography image. The trunk and branches of the blood vessel contour of the second angiography image are defined as second blood vessel regions. The second blood vessel region is divided into a plurality of second blood vessel sub-regions according to the first blood vessel region and in the direction in which the blood vessel extends. The region width of each second blood vessel sub-region is determined so that each second blood vessel region corresponds to a set of region width data.
8. The neurological lesion-assisted identification method according to claim 7, characterized in that: In S5, a cross-Transformer denoising convolutional neural network is used to denoise the third angiography image. A Sobel operator is used to extract edge points from the denoised third angiography image based on image gradient information. Edge points whose distance from the blood vessel contour of the first angiography image is less than a distance threshold are defined as valid edge points. The valid edge points are connected to determine the blood vessel contour of the third angiography image. The trunk and branches of the blood vessel contour of the third angiography image are defined as third blood vessel regions. The third blood vessel region is divided into a plurality of third blood vessel sub-regions according to the first blood vessel region and along the extension direction of the blood vessel. The region width of each third blood vessel sub-region is determined so that each third blood vessel region corresponds to a set of region width data.
9. The neurological lesion auxiliary identification method according to claim 8, characterized in that: In S6, for each first blood vessel region, the corresponding second blood vessel region and third blood vessel region are found, and the region width data of the first blood vessel region is updated according to the corresponding second blood vessel region and third blood vessel region, until the region width data of all first blood vessel regions are updated.
10. The neurological lesion auxiliary identification method according to claim 9, characterized in that: In S7, the change trend of the area width data corresponding to the first blood vessel area is analyzed. If the change trend shows a width reduction trend or a width increase trend, the first blood vessel area is marked as a potential abnormal area. Otherwise, the first blood vessel area is marked as a normal area.