A rapid labeling method for cerebral venous collateral circulation in susceptibility-weighted sequences

Abstract

The present invention relates to the technical field of image analysis, and particularly relates to a method for rapidly annotating cerebral venous collateral circulation in a susceptibility-weighted sequence. Multiple mIP images of a patient's head are acquired and segmented into image patches. Connected regions are obtained based on the gray-scale information of the pixel points in the image patches, and the density of the vascular structure region is determined according to the shape and quantity characteristics of the connected regions. By analyzing the gray-scale distribution, shape, and positional relationship between the connected regions, cerebral venous compensation indicators are determined to evaluate the compensation ability of the cerebral venous structure. A saliency algorithm is adopted, combined with the image patch position, the density of the vascular structure region, and the cerebral venous compensation indicators, to determine the collateral circulation saliency value of each image patch and quantify the saliency degree of the collateral circulation. Finally, a saliency map is obtained through multi-scale fitting, the final saliency value of the pixel points is analyzed, and the regions of interest of the cerebral venous collateral are annotated. By combining various change characteristics of the cerebral venous collateral circulation, the accuracy of the annotation of the cerebral venous structure is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and particularly relates to a method for rapidly annotating cerebral venous collateral circulation in a susceptibility weighted sequence. Background Art

[0002] Susceptibility Weighted Imaging (SWI) enhances the imaging ability of cerebral veins in the brain through the susceptibility effect, can clearly show the change of deoxyhemoglobin concentration, and has important clinical application value in the diagnosis of cerebral infarction and the evaluation of collateral circulation. Cerebral venous collateral circulation (or cerebral venous bypass circulation) mainly provides a blood flow path through the network of cerebral veins and is particularly important in certain brain diseases (such as cerebrovascular diseases, cerebral infarction, etc.).

[0003] Currently, the methods for annotating cerebral venous collateral circulation mainly rely on the automatic annotation technology based on machine vision. However, the existing technologies often only process the gray-scale features of the cerebral vein structure images in SWI imaging to annotate the regions of interest in cerebral venous collateral circulation. However, cerebral venous collateral circulation is a complex physiological process, and the morphological features, gray-scale features, and density of cerebral veins will all change. The processing of a single feature will lead to inaccurate and incomplete final annotation results. Summary of the Invention

[0004] In order to solve the technical problem that cerebral venous collateral circulation is a complex physiological process, and the morphological features, gray-scale features, and density of cerebral veins will all change, and the processing of a single feature will lead to inaccurate and incomplete final annotation results, the purpose of the present invention is to provide a method for rapidly annotating cerebral venous collateral circulation in a susceptibility weighted sequence, and the specific technical solutions adopted are as follows:

[0005] Obtain multiple mIP images of the patient's head;

[0006] At different preset scales, perform region segmentation on each mIP image to obtain image blocks; in each image block, obtain connected regions based on the gray-scale information of pixel points; in each image block, determine the density of the vascular structure region of each image block based on the shape features of the connected regions and the quantity features of pixel points.

[0007] At any one of the preset scales, analyze the gray-scale distribution, shape features, and relative position relationship between the connected regions in each mIP image to determine the cerebral venous compensation index of each connected region.

[0008] In each mIP image, based on the positional relationship between image patches, the density of the vascular structure region in the image patches, and the cerebral venous compensation index of the connected components in the image patches, determine the collateral circulation significance value of each image patch; perform multi-scale fitting on each mIP image based on the significance algorithm to obtain the significance map and the final significance value of each pixel point in the significance map; analyze the final significance value of the pixel points in each significance map to label the region of interest of cerebral venous collateral circulation.

[0009] Further, the method for obtaining the connected components includes:

[0010] Based on the Otsu threshold segmentation algorithm, obtain the binary image of each image patch. In the binary image, regard the pixel points with a gray value of 0 as cerebral venous structure pixel points;

[0011] In each image patch, perform connected component analysis on all cerebral venous structure pixel points to obtain the connected components.

[0012] Further, the method for obtaining the density of the vascular structure region includes:

[0013] In each image patch, after performing convex hull detection on each connected component, use the ratio of the number of pixel points inside the convex hull to the number of pixel points inside the connected component as the quantity weight;

[0014] Use the quantity weight corresponding to the connected component to perform weighted fusion on the number of pixel points inside the convex hull to obtain the density factor corresponding to each image patch;

[0015] The value obtained by normalizing the ratio of the density factor corresponding to each image patch to the number of pixel points inside each image patch is used as the density of the vascular structure region of each image patch.

[0016] Further, the method for obtaining the cerebral venous compensation index includes:

[0017] In each connected component, analyze the gray value fluctuation situation and the shape characteristics of the connected component to obtain the cerebral venous structure index of each connected component;

[0018] In each mIP image, construct a coordinate system and obtain the axis of symmetry. For any image patch, regard the region symmetric about the axis of symmetry in the mIP image as the symmetric patch of the image patch. In the symmetric patch corresponding to each image patch, perform connected component analysis on all cerebral venous structure pixel points to obtain the comparison connected components;

[0019] In each image patch and its corresponding symmetric patch, calculate the Hausdorff distance between any connected component and any comparison connected component, and use the value obtained by performing negative correlation mapping on the minimum value among all Hausdorff distances as the symmetry index of all connected components in each image patch;

[0020] According to the cerebral vein structure index, symmetry index of each connected component, and the average gray value of the pixel points in each connected component, the cerebral vein compensation factor of each connected component is obtained, and the cerebral vein structure index is positively correlated with the cerebral vein compensation factor, and both the symmetry index and the average gray value are negatively correlated with the cerebral vein compensation factor;

[0021] Perform principal component analysis on the position coordinates of all pixel points in each connected component to obtain the main axis. In each image block, take the included angle between the main axes corresponding to any two connected components as the deviation factor, and take the variance of all the deviation factors corresponding to each connected component as the cerebral vein brush eigenvalue;

[0022] Take the value obtained by normalizing the product of the cerebral vein compensation factor corresponding to each connected component and the cerebral vein brush eigenvalue as the cerebral vein compensation index of each connected component.

[0023] Furthermore, the method for obtaining the cerebral vein structure index includes:

[0024] Analyze the shape characteristics of each connected component to obtain the morphological change eigenvalue;

[0025] In each connected component, take the variance of the gray values of all pixel points as the gray value fluctuation eigenvalue;

[0026] Take the value obtained by performing negative correlation mapping and normalization on the product of the morphological change eigenvalue and the gray value fluctuation eigenvalue corresponding to each connected component as the cerebral vein structure index of each connected component.

[0027] Furthermore, the method for obtaining the morphological change eigenvalue includes:

[0028] Obtain the edge line of each connected component and calculate the slope value of each edge pixel point. Take the average value of the absolute values of the differences between the slope values of each edge pixel point and the slope values of the two adjacent edge pixel points as the shape fluctuation factor; take the average value of the shape fluctuation factors between all edge pixel points on the edge line of each connected component as the edge fluctuation eigenvalue;

[0029] For any edge pixel point on the edge line of each connected component, obtain the normal line of the edge line tangent at this edge pixel point, take the other edge pixel point where the normal line intersects the edge line as the corresponding point of this edge pixel point, and take the average value of the Euclidean distances between all edge pixel points and the corresponding points as the shape eigenvalue;

[0030] Take the product of the shape eigenvalue and the edge fluctuation eigenvalue of each connected component as the morphological change eigenvalue.

[0031] Furthermore, the method for obtaining the collateral circulation significant value includes:

[0032] In each mIP image, the Euclidean distance between the centers of any two image patches is used as the distance parameter;

[0033] For any one image patch, taking the other image patches except this image patch as reference patches, the formula model of the collateral circulation significance value includes:

[0034]

[0035] Among them, represents the collateral circulation significance value of each image patch; represents the average value of the cerebral venous compensation indexes of all connected regions in each image patch; represents the th vascular structure region density of the reference patch corresponding to each image patch; represents the distance parameter between each image patch and the th reference patch corresponding to it; represents the exponential function with the natural constant as the base.

[0036] Furthermore, the method for multi-scale fitting of each mIP image based on the significance algorithm to obtain the saliency map and the final saliency value of each pixel point in the saliency map includes:

[0037] At all preset scales, perform multi-scale fitting on each mIP image based on the saliency detection CA algorithm to obtain the saliency map corresponding to each mIP image and the final saliency value of each pixel point in the saliency map.

[0038] Furthermore, the method for obtaining the region of interest of cerebral venous collateral includes:

[0039] In each saliency map, sort the final saliency values of all pixel points in descending order to obtain a sorting sequence;

[0040] In the sorting sequence, calculate the absolute value of the difference between every two adjacent final saliency values as the significance difference value, and take the maximum value among the two final saliency values corresponding to the maximum significance difference value as the reference index;

[0041] Take the pixel points with the final saliency value greater than the reference index as target pixel points;

[0042] In each saliency map, perform morphological processing on all target pixel points to obtain the region of interest of cerebral venous collateral.

[0043] Furthermore, the method for obtaining the image patch includes:

[0044] Use the superpixel segmentation method to perform region segmentation on each mIP image to obtain a number of superpixel patches, and take each superpixel patch as an image patch.

[0045] The present invention has the following beneficial effects:

[0046] Obtain multiple mIP images of the patient's head, perform regional segmentation on each mIP image at different preset scales to obtain image patches, which helps to capture the characteristics of the cerebral venous structure at different scales. Given that the cerebral venous structure usually presents as uniformly distributed black slender blood vessels, in each image patch, connected regions are obtained based on the gray information of pixel points. Moreover, in order to improve the accuracy of subsequent local region comparison, the density of the blood vessel structure region of each image patch can be determined based on the shape characteristics and the number characteristics of pixel points of the connected region. This index can characterize the proportion of the cerebral venous structure in each image patch or the possibility of being the cerebral venous structure, which helps to quantify the distribution of the blood vessel structure. Since most cerebral venous blood vessels are in the shape of smooth lines and are affected by the susceptibility effect, the gray distribution will be relatively uniform. At the same time, when the cerebral venous collateral circulation is activated, the morphological changes of the cerebral venous structure will occur, and then the left-right symmetry characteristics will be destroyed. Therefore, at any preset scale, analyze the gray distribution, shape characteristics, and relative position relationship between the connected regions in each mIP image to determine the cerebral venous compensation index of each connected region. This index can evaluate the compensation ability of the cerebral venous structure. In order to improve the annotation effect, a saliency algorithm is adopted in the present invention. Therefore, in each mIP image, based on the positional relationship between the image patches, the density of the blood vessel structure region of the image patches, and the cerebral venous compensation index of the connected regions in the image patches, the collateral circulation saliency value of each image patch is determined. This step is conducive to quantifying the saliency degree of the collateral circulation and provides a basis for subsequent annotation of the regions of interest of cerebral venous collaterals. Finally, based on the saliency algorithm, multi-scale fitting is performed on each mIP image to obtain a saliency map and the final saliency value of each pixel point in the saliency map, and the final saliency value of the pixel points is analyzed in each saliency map to annotate the regions of interest of cerebral venous collaterals. In summary, the present invention combines various change characteristics in the complex physiological process of cerebral venous collateral circulation, effectively improves the accuracy of segmenting the cerebral venous structure with collateral circulation, and further improves the annotation efficiency of cerebral collateral circulation. Brief Description of the Drawings

[0047] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0048] Figure 1 It is a method flow chart of a method for rapid annotation of cerebral venous collateral circulation in a susceptibility weighted sequence provided by an embodiment of the present invention;

[0049] Figure 2 An mIP image of a patient's head provided by an embodiment of the present invention;

[0050] Figure 3 A schematic diagram of a cerebral venous brush phenomenon provided by an embodiment of the present invention;

[0051] Figure 4 A method flowchart of a method for obtaining cerebral venous collateral circulation indexes provided by an embodiment of the present invention. Detailed implementation manners

[0052] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following combines the accompanying drawings and preferred embodiments to detail the specific implementation manners, structures, features and effects of a method for rapidly annotating cerebral venous collateral circulation in a susceptibility-weighted sequence proposed according to the present invention. In the following description, different "an embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0054] The following specifically describes the specific scheme of a method for rapidly annotating cerebral venous collateral circulation in a susceptibility-weighted sequence provided by the present invention with reference to the accompanying drawings.

[0055] Please refer to Figure 1 , which shows a method flowchart of a method for rapidly annotating cerebral venous collateral circulation in a susceptibility-weighted sequence provided by an embodiment of the present invention, and the method includes the following steps:

[0056] Step S1: Obtain multiple mIP images of the patient's head.

[0057] Susceptibility-weighted imaging (SWI) is an advanced magnetic resonance imaging technique that is highly sensitive to deoxyhemoglobin in cerebral venous blood and can clearly display the cerebral venous structure. Cerebral venous collateral circulation refers to the process in which the cerebral venous system uses alternative paths to assist blood return to maintain local cerebral tissue blood supply and oxygen supply and enable perfusion compensation of ischemic tissue in the case of insufficient arterial blood supply. In clinical diagnosis, the evaluation of cerebral venous collateral circulation is crucial for understanding the pathophysiological process of cerebrovascular diseases; and accurately identifying the cerebral venous collateral region of interest in the image helps doctors more accurately evaluate collateral circulation and judge the patient's condition, etc. Therefore, in this embodiment of the present invention, multiple nuclear magnetic resonance images of the patient's head need to be obtained first.

[0058] Specifically, ensure that the patient is in an appropriate position, usually the supine position. Fix the patient's head with a sponge pad to reduce motion artifacts. Select an appropriate magnetic resonance imaging device (such as a Discovery 750 3.0T MRI scanner). According to clinical needs and the patient's condition, adjust imaging parameters such as magnetic field strength, echo time, repetition time, etc. For example, set the imaging baseline parallel to the anterior commissure-posterior commissure (AC-PC) line to ensure that the scan covers the entire brain; SWI sequence scan parameters: TR 28 ms, TE 20 ms, scan thickness 1 mm, slice gap 1 mm, FOV 220 mm × 193.6 mm, matrix 352 × 352, flip angle 15°, to obtain SWI images. Since cerebral venous vessels usually appear as areas with relatively low signal intensity, i.e., low-density areas, in medical images, and in the annotation of cerebral venous collateral circulation, it is necessary to clearly identify cerebral venous vessels and their branches. Therefore, using minimum intensity projection (mIP) images can effectively reduce the interference caused by brain tissue and other vascular structures to the display of cerebral venous vessels. Thus, with a slice thickness of 20 mm and a slice gap of 1 mm as the reconstruction parameters for volume imaging, mIP images are reconstructed on each obtained SWI image. And for the convenience of subsequent analysis, the mIP images are preprocessed: First, rotate the mIP images to ensure that the left and right brains are in the same left-right direction as the horizontal direction, and convert the DICOM format mIP images to JPG format. After conversion, JPG format mIP images with the same number of layers as the SWI scans are obtained. Each image shows the complete image information of one slice and corresponds to the scan layers. Please refer to Figure 2 , which shows an mIP image of the patient's head in an embodiment of the present invention.

[0059] It should be noted that the acquisition and obtaining of the mIP images of the patient's head in the embodiments of the present invention are all authorized by relevant users, and the process does not violate relevant laws and regulations and does not violate public order and good customs.

[0060] Step S2: At different preset scales, perform regional segmentation on each mIP image to obtain image blocks; in each image block, obtain connected domains based on the gray information of pixel points; in each image block, determine the density of the vascular structure region in each image block based on the shape features and the number features of pixel points of the connected domains.

[0061] Salience detection is a salience detection method that can combine global and local information and is widely used in image segmentation, object detection, and medical image analysis. In the mIP images of SWI sequences, the main features of cerebral venous collateral circulation are the enhanced visualization, dilation, morphological changes, and symmetry changes of cerebral venous structures, which conform to the core concept of salience detection. Therefore, the cerebral venous structures in SWI images where cerebral collateral circulation may appear can be labeled by specifically optimizing the salience detection algorithm.

[0062] The salience detection algorithm identifies and highlights the regions of interest in the image by combining local information and global information. Therefore, in the embodiments of the present invention, by performing region segmentation on the mIP image at different preset scales, multiple image patches are obtained. This multi-scale analysis helps to capture the features of cerebral venous structures at different spatial resolutions, thereby improving the accuracy of recognition. Since cerebral venous structures usually appear as low-signal (black) regions in mIP images, connected regions can be obtained by analyzing the gray-scale information of pixel points. Because cerebral venous structures usually present as uniformly distributed black slender blood vessels, it is usually difficult to separately segment a local region of the cerebral venous structure into an image patch. Therefore, in order to improve the accuracy of subsequent local region comparison, it is necessary to calculate the density of the vascular structure region in each image patch to quantify the distribution characteristics or richness of the cerebral venous structures in the image patch.

[0063] First, at different preset scales, each mIP image is subjected to region segmentation to obtain image patches. In this embodiment of the present invention, the preset scale takes values of , and the specific size of the scale can be adjusted according to the implementation scenario and is not limited here.

[0064] Preferably, in an embodiment of the present invention, the method for obtaining image patches includes:

[0065] Superpixel segmentation can divide an image into a relatively small number of regions (superpixel patches) with similar attributes, and while reducing the image complexity, it can retain the key features of the image, such as edges, textures, and shapes, etc., which has an important impact on the subsequent analysis process. Therefore, in this embodiment of the present invention, the superpixel segmentation method is used to perform region segmentation on each mIP image to obtain a number of superpixel patches, and each superpixel patch is used as an image patch.

[0066] It should be noted that the number of superpixel patches is set to 32, and the specific number can be adjusted according to the implementation scenario and is not limited here; the superpixel segmentation method is a well-known technology, and the specific process is not described in detail here.

[0067] So far, multiple image patches of each mIP image at each preset scale can be obtained. Here, the subsequent analysis process of the mIP image in a single preset scale is described for the convenience of explanation and interpretation.

[0068] After obtaining multiple image patches of each mIP image at each preset scale, connected regions are obtained based on the gray information of pixel points.

[0069] Preferably, in an embodiment of the present invention, the method for obtaining connected regions includes:

[0070] The Otsu threshold segmentation algorithm is an adaptive threshold determination method that can automatically divide an image into a background and an object according to the gray characteristics of the image. Therefore, a binary image of each image patch is obtained based on the Otsu threshold segmentation algorithm. Based on the foregoing analysis, it is known that the cerebral vein structure usually appears as a low-intensity region, that is, a black region. Therefore, in the binary image, pixel points with a gray value of 0 are used as cerebral vein structure pixel points. And in each image patch, connected region analysis is performed on all cerebral vein structure pixel points, so as to obtain connected regions. The connected regions can reflect the shape, size, position and other characteristics of the cerebral vein structure, which is crucial for subsequent recognition, analysis and other processes.

[0071] It should be noted that both the Otsu threshold segmentation algorithm and the connected region analysis are well-known technologies, and the specific processes are not elaborated herein.

[0072] Since in addition to blood vessels, bones contain a large amount of calcium, which will also appear as low-intensity regions in the mIP image, but the cerebral vein structure usually appears as slender blood vessels with uniform distribution and there are bifurcation structures, while the skull usually does not have bifurcation structures. Therefore, in each image patch, based on the shape characteristics and the number characteristics of pixel points of the connected region, the blood vessel structure region density of each image patch is determined to quantify the blood vessel structure density characteristics in the image patch.

[0073] Preferably, in an embodiment of the present invention, the method for obtaining the blood vessel structure region density includes:

[0074] Convex hull detection can extract the outer boundary of the connected region, and this outer boundary can well reflect the shape characteristics of the cerebral vein structure. Therefore, in each image patch, convex hull detection is performed on each connected region, aiming to distinguish different connected regions according to the structural characteristics of the cerebral veins and reduce the interference of bones.

[0075] Since the cerebral vein structure may have bifurcations, the number of pixel points inside the convex hull will increase compared with the number of pixel points in the connected region. Therefore, the ratio of the number of pixel points inside the convex hull to the number of pixel points in the connected region is used as the number weight. The larger the number weight, the higher the possibility of being a cerebral vein structure.

[0076] Then, the number of pixel points within the convex hull is weighted and fused using the quantity weight corresponding to the connected component, to obtain the density factor corresponding to each image block, that is, the quantity weight of each cerebral vein structure is multiplied by the number of pixel points within the corresponding convex hull to obtain a weighted quantity factor, and the sum value of the weighted quantity factors of all connected components is used as the density factor corresponding to the image block. At this time, the larger the density factor, the higher the possibility that the connected component in the image block represents a real cerebral vein structure, and the lower the possibility of bone interference.

[0077] Finally, the value obtained by normalizing the ratio of the density factor corresponding to each image block to the number of pixel points within each image block is used as the density of the vascular structure region of each image block. Normalization can eliminate the influence of the image block size, enabling the densities of the vascular structure regions of all final image blocks to be compared and evaluated within a unified range. At this time, the larger the density of the vascular structure region, the denser the distribution of the cerebral vein blood vessels in the image block, and thus the higher the importance in subsequent saliency analysis. Among them, normalization is a well-known technical means to those skilled in the art. The choice of the normalization function can be linear normalization or standard normalization, etc. The specific normalization method is not limited herein.

[0078] It should be noted that convex hull detection is a well-known technology, and the specific process is not elaborated herein.

[0079] Step S3: At any preset scale, analyze the gray distribution, shape features, and relative position relationship among the connected components in each mIP image, and determine the cerebral vein compensation index of each connected component.

[0080] During the saliency detection process, the calculation of the traditional saliency value is used as the saliency analysis process by comparing the color distance and spatial distance between different image blocks, mainly focusing on high-signal (high-intensity) regions, and usually not involving the morphological features of the structures within the image block. This may reduce the accuracy of the saliency calculation of the cerebral vein structure with collateral circulation. Therefore, in the embodiments of the present invention, the calculation process of the saliency value is optimized.

[0081] The main task is to analyze various features in the image. In the mIP image of the SWI sequence, cerebral veins have unique shape and structural features: cerebral veins are in the shape of smooth lines, and the surface of the blood vessels does not have uneven structures. Compared with other low-intensity regions (such as bones), the connected regions composed of cerebral veins are relatively narrow. Moreover, in normal brain tissue, the blood flow pattern of cerebral veins is stable, and the flow rate of cerebral veins is slower than that of arteries, and the susceptibility effect is more significant. Therefore, the gray-scale distribution of the cerebral vein part in the mIP image of the SWI sequence is relatively uniform. To further improve the calculation accuracy and highlight the low-intensity regions that can represent the collateral circulation of cerebral veins, it is also necessary to analyze the significant values of the cerebral vein compensation features that can be reflected in each connected region. Cerebral veins mainly appear as low intensity on the mIP image of the SWI sequence, with regular morphology and symmetric distribution; when the collateral circulation of cerebral veins is activated, the morphology of cerebral vein structures may change. Due to the increased oxygen consumption in cerebral veins, the concentration of deoxyhemoglobin (dHb) increases, making the cerebral vein region darker (enhanced low signal) on the mIP image. In addition, due to the morphological changes of cerebral veins caused by the activation of the collateral circulation of cerebral veins, the left-right symmetry feature of cerebral veins in the image is destroyed; and when the collateral circulation of cerebral veins compensates, the density of cerebral veins on the ischemic side will increase, forming a brush pattern, which is specifically manifested as: the cerebral vein brush sign shows finer cerebral vein imaging similar to a "brush" that is denser and more disordered than normal cerebral veins (such as Figure 3 the arrow), please refer to Figure 3 , which shows a schematic diagram of the cerebral vein brush phenomenon in an embodiment of the present invention. In summary, the cerebral vein compensation features generated by each connected region can be analyzed through the morphological changes, gray-scale changes, and changes in the symmetry relationship of cerebral veins caused by the collateral circulation of cerebral veins, and the cerebral vein compensation index can be obtained.

[0082] Preferably, in an embodiment of the present invention, the method for obtaining the cerebral vein compensation index includes:

[0083] Please refer to Figure 4 , which shows a flowchart of the method for obtaining the cerebral vein compensation index in an embodiment of the present invention. The method includes the following steps:

[0084] Step S301: In each connected region, analyze the gray-scale fluctuation of pixel points and the shape features of the connected region to obtain the cerebral vein structure index of each connected region.

[0085] Analyze the shape features of each connected region to obtain the morphological change feature value:

[0086] Based on the Canny operator, the edge lines of each connected region are obtained. Then, the least squares method is used to perform curve fitting on the edge lines to obtain the fitted curve and calculate the slope values at each edge pixel point on the fitted curve. The mean value of the absolute differences between the slope values of each edge pixel point and the slope values of its two adjacent edge pixel points is used as the shape fluctuation factor. The larger the shape fluctuation factor, the more tortuous the shape of the edge line of the connected region. The mean value of the shape fluctuation factors between all edge pixel points on the edge line of each connected region is used as the edge fluctuation eigenvalue. The smaller the edge fluctuation eigenvalue, the higher the smoothness of the edge of the connected region, the less it has an uneven structure, and the greater the possibility of being a cerebral venous blood vessel.

[0087] It should be noted that obtaining the edge line based on the Canny operator and performing curve fitting using the least squares method are both well-known techniques, and the specific processes will not be elaborated here.

[0088] Moreover, since cerebral venous blood vessels are usually relatively slender, that is, narrow in width, for any edge pixel point on the edge line of each connected region, the normal line of the tangent line of the edge line is obtained at this edge pixel point, and the other edge pixel point where the normal line intersects the edge line is used as the corresponding point of this edge pixel point. At this time, the distance between the edge pixel points and the corresponding points on the edge line of each connected region can better measure the width characteristics of the blood vessel, and the smaller the distance, the narrower the width. The mean value of the Euclidean distances between all edge pixel points and their corresponding points is used as the shape eigenvalue. The smaller the shape eigenvalue, the smaller the width of the connected region, and the more it conforms to the characteristics of cerebral venous blood vessels. It should be noted that each edge pixel point and its corresponding point are on the edge line of the same connected region. When calculating the Euclidean distance, a coordinate system can be constructed, and the Euclidean distance can be calculated based on the position coordinates between pixel points. The specific process of constructing the coordinate system will not be elaborated here.

[0089] Then, the product of the shape eigenvalue and the edge fluctuation eigenvalue of each connected region is used as the morphological change eigenvalue. Based on the foregoing analysis, the smaller the morphological change eigenvalue, the more the connected region conforms to the characteristics of the cerebral venous structure.

[0090] Because under the influence of the susceptibility effect, the gray-scale distribution inside the cerebral veins is relatively uniform, so in each connected region, the variance of the gray-scale values of all pixel points is used as the gray-scale fluctuation eigenvalue. At this time, the smaller the gray-scale fluctuation eigenvalue, the more uniform the gray-scale distribution of the pixel points, and the more it conforms to the characteristics of the cerebral venous structure.

[0091] Perform a negative correlation mapping and normalization on the product of the morphological change eigenvalue and the gray-scale fluctuation eigenvalue corresponding to each connected component to achieve logical relationship correction, thereby obtaining the cerebral vein structure index for each connected component. That is, the smaller the morphological change eigenvalue of each connected component and the smaller the gray-scale fluctuation eigenvalue, the larger the cerebral vein structure index. Then, the characteristics of the cerebral vein structure possessed by this connected component are more obvious, and it is more likely to represent the area where the cerebral vein structure is located. The negative correlation mapping and normalization here can be carried out using the formula , where represents the exponential function with the natural constant e as the base, and x represents the independent variable.

[0092] Step S302: In each image patch of each mIP image, analyze the symmetry relationship between the image patch and the symmetric region, as well as the gray-scale information within the connected components in the image patch, and fuse the cerebral vein structure indices of the connected components to obtain the cerebral vein compensation factor for each connected component.

[0093] In order to highlight the area that can represent the cerebral vein collateral circulation, it is also necessary to analyze the cerebral vein compensation characteristics that each connected component can reflect.

[0094] When the cerebral vein collateral circulation is activated, it will cause morphological changes in the cerebral veins. Therefore, the characteristic of the left-right symmetry of the cerebral veins in the image will be destroyed. Thus, the symmetry relationship in the image can be analyzed as an index for calculating the cerebral vein compensation factor.

[0095] Construct a coordinate system and obtain the axis of symmetry in each mIP image: Take the lower left corner of the mIP image as the starting point, the horizontal right direction as the positive x-axis direction, and vertically upward at the midpoint of the horizontal axis of the mIP image as the positive y-axis direction to construct a coordinate system. At this time, the y-axis can be used as the axis of symmetry.

[0096] For any image patch, in the mIP image, take the region symmetric about the axis of symmetry as the symmetric patch of this image patch. In the symmetric patch corresponding to each image patch, perform a connected component analysis on all the cerebral vein structure pixel points to obtain a comparison connected component.

[0097] The Hausdorff distance is a set metric for measuring the similarity between two sets. In image processing, it can evaluate the similarity between two regions or shapes, and further determine their symmetry relationship.

[0098] Therefore, in each image block and its corresponding symmetric block, calculate the Hausdorff distance between any connected component and any comparison connected component. The larger the Hausdorff distance, the lower the similarity. The value obtained by performing a negative correlation mapping on the minimum value among all Hausdorff distances is used as the symmetry index of all connected components in each image block. Performing a negative correlation mapping on the minimum value of the Hausdorff distance can correct the logical relationship. At this time, the larger the symmetry index, the higher the similarity, which is regarded as better symmetry, and the smaller the degree of cerebral venous compensation; conversely, the smaller the symmetry index, the lower the similarity, which is regarded as worse symmetry, and the larger the degree of cerebral venous compensation.

[0099] When the cerebral venous collateral circulation is activated, the oxygen consumption of cerebral veins increases. Therefore, the cerebral venous region becomes darker on the mIP image, so the gray value can also be used as an index to measure cerebral venous compensation. Therefore, according to the cerebral venous structure index, symmetry index of each connected component, and the average gray value of pixel points in each connected component, the cerebral venous compensation factor of each connected component is obtained. The formula model of the cerebral venous compensation factor includes:

[0100]

[0101] Among them, represents the cerebral venous compensation factor of each connected component; represents the cerebral venous structure index of each connected component; represents the average gray value of pixel points in each connected component; represents the symmetry index of each connected component; represents a preset first parameter; represents a normalization function.

[0102] In the formula model of the cerebral venous compensation factor, based on the foregoing analysis, it can be seen that the smaller the symmetry index, the more the left-right symmetry of the veins in the image is damaged, and the greater the degree of cerebral venous compensation; the smaller the average gray value, the more the cerebral venous collateral circulation is activated, which is also regarded as the greater the degree of cerebral venous compensation; and the larger the cerebral venous structure index, the more obvious the characteristics of the cerebral venous structure possessed by this connected component. Therefore, the cerebral venous structure index is used as a weight here to adjust the prominence with other connected components. Therefore, the cerebral venous structure index is positively correlated with the cerebral venous compensation factor, and both the symmetry index and the average gray value are negatively correlated with the cerebral venous compensation factor. Therefore, the above formula model is constructed based on this logic to calculate the cerebral venous compensation factor. At this time, the larger the cerebral venous compensation factor, the higher the significance of the cerebral venous compensation characteristics of the connected component.

[0103] It should be noted that the preset first parameter is used to prevent the denominator from being zero, and can take the value of 0.001. The specific value can be adjusted according to the implementation scenario and is not limited here.

[0104] Step S303: In each image block, analyze the positional relationship between the connected components, and determine the cerebral vein brush eigenvalue of each connected component.

[0105] When there is collateral circulation compensation in the cerebral veins, there will be denser and more disordered small vein imaging similar to "brushes". Therefore, the morphological change characteristics caused by cerebral vein collateral circulation can also be analyzed to obtain the cerebral vein brush eigenvalue, which is used to reflect the compensation situation of the cerebral veins.

[0106] Perform principal component analysis on the position coordinates of all pixel points in each connected component to obtain the main axis. This main axis represents the main extension direction of the connected component and is an important feature for analyzing the positional relationship between connected components. In each image block, the angle between the main axes corresponding to any two connected components is used as the deviation factor. The larger the deviation factor, the greater the positional deviation between the two connected components. Finally, the variance of all deviation factors corresponding to each connected component is used as the cerebral vein brush eigenvalue. The larger the cerebral vein brush eigenvalue at this time, the greater the degree of disorder in the positional distribution between the connected components, and then it can be regarded as the more significant the characteristics of cerebral vein collateral circulation compensation.

[0107] Step S304: Integrate the cerebral vein compensation factor and the cerebral vein brush eigenvalue of each connected component to obtain the cerebral vein compensation index of each connected component.

[0108] Based on the analysis of the foregoing steps, it can be seen that the larger the cerebral vein compensation factor of a certain connected component, the higher the significance of the cerebral vein compensation characteristics of this connected component; similarly, the larger the cerebral vein brush eigenvalue of a certain connected component, the greater the degree of disorder in the positional distribution between this connected component, and then it can be regarded as the more significant the characteristics of its cerebral vein collateral circulation compensation. Therefore, the value obtained by normalizing the product of the cerebral vein compensation factor and the cerebral vein brush eigenvalue corresponding to each connected component is used as the cerebral vein compensation index of each connected component. At this time, the larger the cerebral vein compensation index, the more significant the cerebral vein compensation characteristics of this connected component, and then the higher the attention. Among them, normalization is a technical means well-known to those skilled in the art. The selection of the normalization function can be linear normalization or standard normalization, etc. The specific normalization method is not limited here.

[0109] Step S4: In each mIP image, based on the positional relationship between the image blocks, the density of the vascular structure area of the image blocks, and the cerebral vein compensation index of the connected components in the image blocks, determine the collateral circulation significance value of each image block; perform multi-scale fitting on each mIP image based on the significance algorithm to obtain the significance map and the final significance value of each pixel point in the significance map; analyze the final significance value of the pixel points in each significance map to mark the cerebral vein collateral interest area.

[0110] In the process of saliency detection, the traditional saliency value is calculated by comparing the color distance and position distance between different image patches. In the embodiments of the present invention, in order to improve the subsequent labeling effect of cerebral venous collateral circulation, the calculation process of the saliency value is adjusted by analyzing the positional relationship between image patches, the density of the vascular structure region of the image patches, and the cerebral venous compensation index of the connected regions in the image patches, and the collateral circulation saliency value of each image patch is obtained.

[0111] Preferably, in one embodiment of the present invention, the method for obtaining the collateral circulation saliency value includes:

[0112] In each mIP image, the Euclidean distance between the centers of any two image patches is used as the distance parameter;

[0113] For any one image patch, taking the other image patches except this image patch as reference patches, the formula model of the collateral circulation saliency value includes:

[0114]

[0115] Wherein, represents the collateral circulation saliency value of each image patch; represents the average value of the cerebral venous compensation indexes of all connected regions in each image patch; represents the th reference patch corresponding to each image patch, and is the density of the vascular structure region; represents the distance parameter between each image patch and the th reference patch corresponding thereto; represents the exponential function with the natural constant as the base.

[0116] In the formula model of the collateral circulation saliency value, for each image patch, the distance parameter between it and the reference patch is calculated, and the weight of the reference patch with a higher vascular structure density in the saliency comparison with the image patch is increased through the index of the density of the vascular structure region, improving the comparison effect; and the comparison effect is adjusted by using the average value of the cerebral venous compensation indexes of all connected regions in each image patch, which can further amplify the calculation result of the saliency of the venous structure with the characteristics of cerebral venous collateral circulation. Thus, the collateral circulation saliency value of each image patch is obtained, that is, the collateral circulation saliency value of all pixel points in each image patch is obtained, and the collateral circulation saliency values of the pixel points in each graphic patch are the same.

[0117] It should be noted that the formula model of the collateral circulation saliency value is constructed with reference to the formula model of the saliency value in the saliency detection CA algorithm, so the construction process of the formula model will not be elaborated here.

[0118] So far, based on the above process, the collateral circulation significance values of each image patch at each preset scale can be obtained. To enhance the contrast between the significant region and the non-significant region, multi-scale fitting can be performed on each mIP image based on the significance algorithm, so as to obtain the significant map and the final significance value of each pixel point in the significant map.

[0119] Preferably, in an embodiment of the present invention, multi-scale fitting is performed on each mIP image based on the significance algorithm to obtain the significant map and the final significance value of each pixel point in the significant map, including:

[0120] At all preset scales, multi-scale fitting is performed on each mIP image based on the significance detection CA algorithm to obtain the significant map corresponding to each mIP image and the final significance value of each pixel point in the significant map.

[0121] It should be noted that the process of the significance detection CA algorithm for obtaining the significant map and the final significance value is a well-known technology, and the specific process will not be elaborated here.

[0122] After obtaining the significant map, the region of interest of the cerebral venous collateral can be determined based on the final significance value of each pixel point in the significant map, and then it can be labeled.

[0123] Preferably, in an embodiment of the present invention, the method for obtaining the region of interest of the cerebral venous collateral includes:

[0124] In each significant map, the final significance values of all pixel points are sorted in descending order to obtain a sorting sequence.

[0125] In the sorting sequence, the absolute value of the difference between every two adjacent final significance values is calculated as the significance difference value. This process can identify the mutation points between the final significance values, and then the maximum value among the two final significance values corresponding to the maximum significance difference value is used as a reference index. The reference index provides a threshold that can be used to distinguish significant and non-significant pixel points.

[0126] The pixel points with final significance values greater than the reference index are used as target pixel points, and in each significant map, morphological processing is performed on all the target pixel points to eliminate internal voids, and the connected domain obtained after processing is used as the region of interest of the cerebral venous collateral.

[0127] Finally, on each mIP image in the SWI sequence, the cerebral venous collateral circulation venous structure is marked according to the marked region of interest of the cerebral venous collateral to complete the annotation process.

[0128] In summary, obtaining multiple mIP images of a patient's head, performing regional segmentation on each mIP image at different preset scales to obtain image patches, helps capture the characteristics of cerebral venous structures at different scales. Given that cerebral venous structures usually present as uniformly distributed black slender blood vessels, in each image patch, cerebral venous structure pixel points are marked based on the gray information of pixel points and connected components are obtained. Moreover, to improve the accuracy of subsequent local region comparison, the density of the vascular structure region of each image patch can be determined based on the shape characteristics and the number characteristics of pixel points of the connected components. This index can characterize the proportion of the cerebral venous structure in each image patch or the possibility of the cerebral venous structure, and helps quantify the distribution of the vascular structure. Since most cerebral venous blood vessels are in the shape of smooth lines and are affected by the susceptibility effect, the gray distribution will be relatively uniform. At the same time, when the cerebral venous collateral circulation is activated, the morphological changes of the cerebral venous structure will occur, and then the left-right symmetry feature will be destroyed. Therefore, at any preset scale, analyze the gray distribution, shape characteristics, and relative position relationship between connected components in each mIP image to determine the cerebral venous compensation index of each connected component. This index can evaluate the compensation ability of the cerebral venous structure. To improve the annotation effect, a saliency algorithm is adopted in the embodiments of the present invention. Therefore, in each mIP image, based on the positional relationship between image patches, the density of the vascular structure region of the image patch, and the cerebral venous compensation index of the connected components in the image patch, the collateral circulation saliency value of each image patch is determined. This step is beneficial to quantify the saliency degree of the collateral circulation and provides a basis for subsequent annotation of cerebral venous collateral regions of interest. Finally, based on the saliency algorithm, multi-scale fitting is performed on each mIP image to obtain a saliency map and the final saliency value of each pixel point in the saliency map, and analyze the final saliency value of the pixel points in each saliency map to annotate the cerebral venous collateral regions of interest. In summary, the embodiments of the present invention combine various change characteristics in the complex physiological process of cerebral venous collateral circulation, effectively improve the accuracy of segmenting cerebral venous structures with collateral circulation, and further improve the efficiency of cerebral collateral circulation annotation.

[0129] It should be noted that: the above sequence of the embodiments of the present invention is only for description and does not represent the advantages or disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0130] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for rapidly annotating the collateral circulation of cerebral veins in a susceptibility-weighted sequence, characterized in that, The method includes: Obtaining multiple mIP images of a patient's head; At different preset scales, performing region segmentation on each mIP image to obtain image patches; in each image patch, obtaining connected components based on the gray information of pixel points; in each image patch, determining the density of the vascular structure region of each image patch based on the shape characteristics and the number characteristics of the connected components; At any one preset scale, analyzing the gray distribution, shape characteristics, and relative position relationship between the connected components in each mIP image to determine the cerebral venous compensation index of each connected component; In each mIP image, based on the positional relationship between the image patches, the density of the vascular structure region of the image patches, and the cerebral venous compensation index of the connected components in the image patches, determining the collateral circulation significance value of each image patch; performing multi-scale fitting on each mIP image based on the significance algorithm to obtain a significance map and the final significance value of each pixel point in the significance map; analyzing the final significance value of the pixel points in each significance map to label the region of interest of the cerebral venous collateral.

2. The rapid labeling method for collateral circulation of cerebral veins in a susceptibility-weighted sequence according to claim 1, wherein The method for obtaining the connected components includes: Obtaining a binary image of each image patch based on the Otsu threshold segmentation algorithm, and in the binary image, taking the pixel points with a gray value of 0 as cerebral venous structure pixel points; In each image patch, performing connected component analysis on all the cerebral venous structure pixel points to obtain the connected components.

3. A method for rapidly labeling the collateral circulation of the middle cerebral vein in a susceptibility-weighted sequence according to claim 1, characterized in that, The method for obtaining the density of the vascular structure region includes: In each image patch, after performing convex hull detection on each connected component, taking the ratio of the number of pixel points inside the convex hull to the number of pixel points inside the connected component as the number weight; Using the number weight corresponding to the connected component to perform weighted fusion on the number of pixel points inside the convex hull to obtain the density factor corresponding to each image patch; Taking the value obtained by normalizing the ratio of the density factor corresponding to each image patch to the number of pixel points in each image patch as the density of the vascular structure region of each image patch.

4. A method for rapidly labeling the collateral circulation of the cerebral veins in a susceptibility-weighted sequence according to claim 2, characterized in that The method for obtaining the cerebral venous compensation index includes: In each connected component, analyzing the gray fluctuation situation of the pixel points and the shape characteristics of the connected component to obtain the cerebral venous structure index of each connected component; Constructing a coordinate system and obtaining the axis of symmetry in each mIP image. For any one image patch, taking the region symmetric about the axis of symmetry in the mIP image as the symmetric patch of the image patch, and in the symmetric patch corresponding to each image patch, performing connected component analysis on all the cerebral venous structure pixel points to obtain the comparison connected components; In each image patch and its corresponding symmetric patch, calculating the Hausdorff distance between any one connected component and any one comparison connected component, and taking the value obtained by performing negative correlation mapping on the minimum value of all the Hausdorff distances as the symmetry index of all the connected components in each image patch; According to the cerebral venous structure index, symmetry index of each connected component, and the average gray value of the pixel points in each connected component, obtaining the cerebral venous compensation factor of each connected component, and the cerebral venous structure index is positively correlated with the cerebral venous compensation factor, and both the symmetry index and the average gray value are negatively correlated with the cerebral venous compensation factor; Perform principal component analysis on the position coordinates of all pixel points in each connected component to obtain the main axis. In each image patch, take the angle between the main axes corresponding to any two connected components as the deviation factor, and take the variance of all deviation factors corresponding to each connected component as the cerebral vein brush eigenvalue; Take the value obtained by normalizing the product of the cerebral vein compensation factor and the cerebral vein brush eigenvalue corresponding to each connected component as the cerebral vein compensation index for each connected component.

5. A method for rapidly annotating the collateral circulation of the cerebral veins in a susceptibility-weighted sequence according to claim 4, characterized in that, The method for obtaining the cerebral vein structure index includes: Analyze the shape characteristics of each connected component to obtain the morphological change eigenvalue; In each connected component, take the variance of the gray values of all pixel points as the gray fluctuation eigenvalue; Take the value obtained by performing negative correlation mapping and normalization on the product of the morphological change eigenvalue and the gray fluctuation eigenvalue corresponding to each connected component as the cerebral vein structure index for each connected component.

6. A method for rapidly labeling the collateral circulation of the cerebral veins in a susceptibility-weighted sequence according to claim 5, characterized in that The method for obtaining the morphological change eigenvalue includes: Obtain the edge line of each connected component and calculate the slope value of each edge pixel point. Take the mean value of the absolute values of the differences between the slope values of each edge pixel point and the adjacent two edge pixel points as the shape fluctuation factor; take the mean value of the shape fluctuation factors between all edge pixel points on the edge line of each connected component as the edge fluctuation eigenvalue; For any edge pixel point on the edge line of each connected component, obtain the normal line of the edge line tangent at this edge pixel point, take the other edge pixel point where the normal line intersects the edge line as the corresponding point of this edge pixel point, and take the mean value of the Euclidean distances between all edge pixel points and the corresponding points as the shape eigenvalue; Take the product of the shape eigenvalue and the edge fluctuation eigenvalue of each connected component as the morphological change eigenvalue.

7. A method for rapidly labeling the collateral circulation of the middle cerebral vein in a susceptibility-weighted sequence according to claim 1, characterized in that The method for obtaining the collateral circulation significance value includes: In each mIP image, take the Euclidean distance between the centers of any two image patches as the distance parameter; For any one image patch, take the other image patches except this image patch as reference patches. The formula model of the collateral circulation significance value includes: Among them, represents the collateral circulation significant value of each image block; represents the mean value of the cerebral vein compensation index of all connected regions in each image block; represents the th reference block corresponding to each image block in terms of the vascular structure area density; represents the distance parameter between each image block and the th reference block corresponding thereto; represents the exponential function with the natural constant as the base.

8. A method for rapidly annotating the collateral circulation of the cerebral veins in a susceptibility-weighted sequence according to claim 1, characterized in that The multi-scale fitting of each mIP image based on the saliency algorithm to obtain the saliency map and the final saliency value of each pixel point in the saliency map includes: At all preset scales, perform multi-scale fitting on each mIP image based on the saliency detection CA algorithm to obtain the saliency map corresponding to each mIP image and the final saliency value of each pixel point in the saliency map.

9. A method for rapidly labeling the collateral circulation of the cerebral veins in a susceptibility-weighted sequence according to claim 1, characterized in that, The method for obtaining the cerebral vein collateral region of interest includes: In each saliency map, sort the final saliency values of all pixel points in descending order to obtain a sorted sequence; In the sorted sequence, calculate the absolute value of the difference between every two adjacent final saliency values as the saliency difference value, and take the maximum value among the two final saliency values corresponding to the maximum saliency difference value as the reference index; Take the pixel points with final saliency values greater than the reference index as target pixel points; In each saliency map, perform morphological processing on all target pixel points to obtain the cerebral vein collateral region of interest.

10. A method for rapidly labeling the collateral circulation of the middle cerebral vein in a susceptibility-weighted sequence according to claim 1, characterized in that The method for obtaining the image patch includes: Use the superpixel segmentation method to perform regional segmentation on each mIP image to obtain a number of superpixel blocks, and take each superpixel block as an image patch.

Images