CN117011619A - Method and device for identifying coronary artery branches in XRA images - Google Patents
Method and device for identifying coronary artery branches in XRA images Download PDFInfo
- Publication number
- CN117011619A CN117011619A CN202310602873.9A CN202310602873A CN117011619A CN 117011619 A CN117011619 A CN 117011619A CN 202310602873 A CN202310602873 A CN 202310602873A CN 117011619 A CN117011619 A CN 117011619A
- Authority
- CN
- China
- Prior art keywords
- coronary
- coronary artery
- branch
- branches
- vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/7715—Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/20—Drawing from basic elements, e.g. lines or circles
- G06T11/203—Drawing of straight lines or curves
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/776—Validation; Performance evaluation
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20112—Image segmentation details
- G06T2207/20132—Image cropping
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30101—Blood vessel; Artery; Vein; Vascular
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Medical Informatics (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Image Analysis (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The coronary artery branch identification method and device of the XRA image can quickly and accurately realize branch identification of the coronary artery XRA image. Comprising the following steps: (1) rapid pre-segmentation of coronary artery adaptive parameters; (2) Drawing a coronary artery segment path by a rapid relay travel optimization method; (3) constructing a multi-mode coronary topological graph by traversing a rejection method; (4) extracting the characteristics of the azimuth vector field inside the coronary artery segment; (5) angular guidance and proportional fusion of three-dimensional coronary features; (6) Three types of coronary features are fused together by a joint graph attention network structure.
Description
Technical Field
The invention relates to the technical field of visual image processing, in particular to a coronary artery branch identification method of an XRA image and a coronary artery branch identification device of the XRA image.
Background
Coronary artery disease (Coronary Artery Diseases, CAD), caused by coronary artery stenosis or occlusion, is one of the most common cardiovascular diseases. The occurrence of lesions in different coronary branches can bring different influences to the disease, so in clinical practice, doctors need to distinguish lesions in different branches of coronary branches to diagnose the disease, however, the coronary branches have complex structures and are difficult to automatically identify, and difficulties are brought to CAD diagnosis and subsequent treatment.
In the aspect of the existing coronary artery branch identification method, the traditional method generally builds a coronary artery blood vessel model, main branches of a blood vessel tree are determined by a registration method to subdivide categories of other branches, but because of complex coronary artery topological structure and large individual difference, the method is difficult to build a universal blood vessel model, and meanwhile, the traditional algorithm has poor robustness and relies on manual experience, so that the operation is complex. In order to improve the problem, the pixel-level semantic segmentation method judges each pixel type of the coronary image in a data driving mode, but the imaging conditions are different, so that the problems of uneven gray scale, low contrast and artifact of the coronary image are caused, and the gray scale feature expression capability is not strong. Meanwhile, in terms of imaging data, XRA (X Ray images) images have the problems of overlapping coronary branches and lack of spatial information, which leads to branch misrecognition. These problems present a great challenge to the identification of coronary branches.
Disclosure of Invention
In order to overcome the defects of the prior art, the technical problem to be solved by the invention is to provide a coronary artery branch identification method of an XRA image, which can quickly and accurately realize branch identification of the coronary artery XRA image.
The technical scheme of the invention is as follows: the coronary artery branch identification method of the XRA image comprises the following steps:
(1) The coronary artery adaptive parameter is rapidly pre-segmented, an adaptive framework nnUnet based on 2D U-Net is used, depth separable convolution is adopted to replace traditional convolution, the parameter quantity is reduced under the condition that the segmentation precision is unchanged, and the reasoning speed is increased; in preprocessing, in order to reduce the amount of computation, the framework uses clipping preprocessing for the zero-value regions;
(2) Drawing a coronary artery segment path by a rapid relay travel optimization method, and extracting a coronary artery central line by a semi-automatic method based on travel optimization;
(3) Constructing a multi-mode coronary artery topological graph by using a traversal elimination method as a core, completing automatic identification of a coronary artery central line endpoint, identifying branch boundaries and resampling central line points to enable the number of the branch boundaries to be consistent, constructing the coronary artery topological graph, cutting the consistent coronary artery central line points into identifiable branches, and obtaining nodes and edges required in the topological graph;
(4) Extracting features of internal azimuth vector fields of coronary artery segments, defining internal azimuth vector fields to simulate the dependence of different coronary artery branches on other internal branches, evaluating the relation between different coronary artery branches by respectively calculating the relative position and growth direction of each coronary artery branch, cascading the feature information together to complete the construction of the internal azimuth vector fields of the coronary artery after calculating the features of the relative distance, direction and growth direction of all the branches of the coronary artery, and providing node features for a coronary artery topological graph;
(5) Angle guiding and proportion fusion of three-dimensional coronary artery characteristics are carried out, an XRA image is divided into four visual angles to respectively observe different branches, 3D coronary artery branch recognition results are utilized to guide 2D characteristics, angle guiding characteristics are extracted, then 3D characteristics and 2D characteristics are fused to obtain fusion characteristics, and missing space information is fully supplemented;
(6) The three types of coronary artery characteristics are fused together through a joint graph attention network structure, the characteristic expression capacity is improved, the joint graph attention network consists of two layers of graph attention, namely a node layer and a joint layer, the weight coefficient of a progressive automatic learning neighborhood node to a central node is obtained through weighted aggregation, the node layer learns the weight of the neighborhood node in each group of coronary artery topological graph respectively, the joint graph nodes are calculated through aggregation, the three joint graph nodes respectively represent different types of coronary artery characteristic information, the joint layer extracts the correlation relation of the three joint graph nodes, the importance degree of the three branch information to the branch class is measured, and the branch identification result is finally weighted aggregated.
According to the invention, through a two-layer graph attention network, the importance of 2D azimuth vector field features, angle guiding features and fusion features to branch recognition tasks is progressively measured, the weight coefficients of the two-layer graph attention network and the angle guiding features and the fusion features are calculated, and finally three types of feature information are fused together in a weighted aggregation mode, so that the expression capability of 2D coronary artery features is enhanced, the coronary artery branch recognition effect of an XRA image is improved, and the branch recognition of the coronary artery XRA image can be rapidly and accurately realized.
Also provided is a coronary branch identification device of an XRA image, comprising:
the coronary artery adaptive parameter rapid pre-segmentation module is configured to use an adaptive framework nnUnet based on 2D U-Net, adopts depth separable convolution to replace traditional convolution, reduces the parameter quantity under the condition of unchanged segmentation precision, and quickens the reasoning speed; in preprocessing, in order to reduce the amount of computation, the framework uses clipping preprocessing for the zero-value regions;
the rapid relay travel optimization method draws a coronary artery segment path module configured to extract a coronary artery centerline using a semi-automatic approach based on travel optimization;
constructing a multi-mode coronary artery topological graph module by using a traversal elimination method, wherein the module is configured to take the traversal elimination method as a core, finish automatic identification of a coronary artery central line endpoint, identify branch boundaries and resample central line points to enable the number of the branch boundaries to be consistent, construct a coronary artery topological graph, intercept the consistent coronary artery central line points into identifiable branches, and acquire nodes and edges required in the topological graph;
the internal azimuth vector field feature extraction module of the coronary segment is configured to define an internal azimuth vector field to simulate the dependence of different coronary branches on other internal branches, evaluate the relation among different coronary branches by respectively calculating the relative position and the growth direction of each coronary branch, and after calculating the features of the relative distance, the direction and the growth direction of all the branches of the coronary, cascade the feature information together to complete the construction of the internal azimuth vector field of the coronary, which is the coronary
The pulse topology map provides node features;
the angle guiding and proportion fusing module of the three-dimensional coronary artery features is configured to divide the XRA image into four visual angles to respectively observe different branches, guide the 2D features by using the 3D coronary artery branch recognition result, extract the angle guiding features, then fuse the 3D features and the 2D features to obtain fused features, and fully supplement the missing space information;
the coronary feature fusion module is configured to fuse three types of coronary features together through a joint graph attention network structure, improve feature expression capacity, the joint graph attention network is composed of two layers of graph attention, namely a node layer and a joint layer, the weight coefficient of a progressive automatic learning neighborhood node to a central node is obtained through weighted aggregation, the node layer learns the weight of the neighborhood node in each group of coronary topological graph respectively, the joint graph nodes are calculated through aggregation, the three joint graph nodes respectively represent different types of coronary feature information, the joint layer extracts the correlation of the three joint graph nodes, the importance degree of the three branch information to the branch class is measured, and the branch identification result is finally weighted aggregated.
Drawings
FIG. 1 is a flow chart of a method of coronary branch identification of an XRA image according to the invention.
Detailed Description
As shown in fig. 1, this method of coronary branch identification for XRA images includes the steps of:
(1) The coronary artery adaptive parameter is rapidly pre-segmented, an adaptive framework nnUnet based on 2D U-Net is used, depth separable convolution is adopted to replace traditional convolution, the parameter quantity is reduced under the condition that the segmentation precision is unchanged, and the reasoning speed is increased; in preprocessing, in order to reduce the amount of computation, the framework uses clipping preprocessing for the zero-value regions;
(2) Drawing a coronary artery segment path by a rapid relay travel optimization method, and extracting a coronary artery central line by a semi-automatic method based on travel optimization;
(3) Constructing a multi-mode coronary artery topological graph by using a traversal elimination method as a core, completing automatic identification of a coronary artery central line endpoint, identifying branch boundaries and resampling central line points to enable the number of the branch boundaries to be consistent, constructing the coronary artery topological graph, cutting the consistent coronary artery central line points into identifiable branches, and obtaining nodes and edges required in the topological graph;
(4) Extracting features of internal azimuth vector fields of coronary artery segments, defining internal azimuth vector fields to simulate the dependence of different coronary artery branches on other internal branches, evaluating the relation between different coronary artery branches by respectively calculating the relative position and growth direction of each coronary artery branch, cascading the feature information together to complete the construction of the internal azimuth vector fields of the coronary artery after calculating the features of the relative distance, direction and growth direction of all the branches of the coronary artery, and providing node features for a coronary artery topological graph;
(5) Angle guiding and proportion fusing of three-dimensional coronary artery characteristics, dividing an XRA image into four visual angles to respectively observe different branches, guiding 2D characteristics by using 3D coronary artery branch identification results,
extracting angle guiding features, fusing 3D and 2D features to obtain fused features,
the missing space information is fully supplemented;
(6) The three types of coronary artery characteristics are fused together through a joint graph attention network structure, the characteristic expression capability is improved, the joint graph attention network consists of two layers of graph attention, which are respectively called a node layer and a joint layer, the weight coefficient of a neighborhood node to a central node is automatically learned in a progressive manner, finally, an identification result is obtained through weighted aggregation, the node layer respectively learns the weight of the neighborhood node in each group of coronary artery topological graph, the joint graph node is calculated through aggregation,
the three joint graph nodes respectively represent different types of coronary artery characteristic information, the joint layer extracts the correlation relation of the three joint graph nodes, the importance degree of the three branch information on the branch categories is measured, and finally the branch recognition results are weighted and aggregated.
According to the invention, through a two-layer graph attention network, the importance of 2D azimuth vector field features, angle guiding features and fusion features to branch recognition tasks is progressively measured, the weight coefficients of the two-layer graph attention network and the angle guiding features and the fusion features are calculated, and finally three types of feature information are fused together in a weighted aggregation mode, so that the expression capability of 2D coronary artery features is enhanced, the coronary artery branch recognition effect of an XRA image is improved, and the branch recognition of the coronary artery XRA image can be rapidly and accurately realized.
Preferably, in the step (1), in terms of network, the U-Net network structure includes an encoder, a jump connection, and a decoder, where the encoder is composed of a depth separable convolution layer, a leakage ReLU activation function, instance Normalization, and a 2×2 max pooling layer, and the U-Net extracts low-level and high-level semantic feature graphs of the coronary image, and fuses semantic features of different scales; the convolution layer adopts depth separable convolution, the convolution calculation is divided into two steps, the number of convolution kernels is the same as that of the channels of the previous layer, the channels and the convolution kernels are in one-to-one correspondence, the size of the convolution kernels is 1 multiplied by M, M is the number of the channels of the previous layer, and the point-to-point convolution operation carries out weighted combination on the feature map of the first step in the depth direction to generate a new feature map;
in the aspect of network training, the training super-parameters are adaptively adjusted according to the size of the coronary artery image, and the frame nnU-Net monitors the occupation condition of the video memory in real time and automatically sets the batch size and the image block size to prevent the explosion of the video memory.
Preferably, in the step (1), the adjustment of the Batch size is prioritized over the Batch; the size of Patch ensures that at least one third of the sampled area is coronary; using the sum of the cross entropy loss and the Dice loss as a loss function; the learning rate is adaptively adjusted according to the exponential average of the loss function.
The Fast Marching and shortest Path (Minimum Path) method can realize image linear feature analysis and can be applied to coronary structure topological graph extraction. The method utilizes the image characteristic difference of the coronary blood vessel and the background to construct an energy diagram, wherein in the energy diagram, the energy value at the center of the coronary blood vessel is smaller, and the energy value towards the blood vessel edge and the background direction is gradually increased. In the semiautomatic method of this section, a starting point is manually set, and the minimum energy map corresponding to the point is obtained by solving the Eikonal equation by using a fast-travelling method on the velocity map. And setting an ending point, reversely searching the starting point to minimize the energy of the path, and obtaining the shortest path which can be used as the central line of the coronary artery. Preferably, in the step (2), the branch end points and the bifurcation points of the coronary artery of the concerned part are manually determined as key points, and the shortest path search drawing among a plurality of key points is realized by using an improved travelling optimization method; constructing an energy map by utilizing the image characteristic difference of the coronary blood vessel and the background, wherein in the energy map, the energy value at the center of the coronary blood vessel is smaller, and the energy value towards the edge of the blood vessel and the background direction is gradually increased; manually setting a starting point, solving an Eikonal equation on a speed map by using a rapid advancing method to obtain a minimum energy map corresponding to the point, setting a termination point, reversely searching the starting point to enable the energy of the path to be minimum, and obtaining a shortest path which is used as the central line of the coronary artery.
Preferably, in said step (2), regarding the solving of the energy map, a coronary image I is first defined in which the pixel values of the centerline position have smaller values, assuming two points p on the map 1 、p 2 ，B p Is p 1 、p 2 A set of paths between, defining an energy map function E:the expression is as follows:
E(δ)＝∫I(δ(z))dz (2.1)
where δ is a path between two points and z is a length parameter. The delta that minimizes the energy E (delta) is the coronary centerline between the two points; when searching the center line, find the corresponding p 1 Is defined by the minimum energy diagram phi:any point on the image +.>The value in the minimum energy diagram is p 1 To->Energy value corresponding to coronary centerline:
satisfies the Eikonal equation:
the fast travelling method is a numerical solution for Eikonal equation, a stable solution for Eikonal equation is solved by using a reverse difference method, a first-order forward and backward difference operator is used for replacing a differential quotient, and the calculation of a single pixel point is as follows:
wherein h is x And h y Representing the pixel difference size in the x, y direction of the image.
The fast-travelling algorithm does not need redundant priori knowledge and post-processing, only needs to give a starting point and an ending point, but in the extraction of the coronary topological graph, each branch needs to be manually marked with the starting point and the ending point because of more branches of the coronary artery, so that the fast-travelling algorithm is excessively complicated, and meanwhile, the coronary artery has the characteristics of a single tree structure and good connectivity. Therefore, the relay point optimization fast traveling algorithm is used in the section, only one time of energy diagram calculation is needed, the calculation efficiency is improved, and the possible false shortcuts can be avoided while the manual marking times are reduced.
Because only one coronary tree exists in the coronary image, and the connectivity of the coronary is good, the energy diagram corresponding to a starting point in the coronary is theoretically calculated, and then one ending point is arbitrarily selected, the shortest path can be reversely searched. In order to avoid wrong shortcuts, relay points can be added to complete multiple reverse searches, and finally, starting points are searched. Thus, if the coronary image has not yet calculated an energy map, then an energy map corresponding to the first keypoint of the queue is calculated and the coronary pre-segmentation result is superimposed as a global minimum energy map Φ. The specific solving steps are shown in the pseudo code of table 1.
TABLE 1
After manual selection of the coronary endpoint and branch intersection, these keypoints are stored in a queue. And when solving the energy map, taking the pixel point with the energy calculated as an initial point each time, updating the minimum energy of the pixel points in the four adjacent domains according to the energy of the point, and finally obtaining the minimum energy map of the first key point of the queue by circulating the steps.
And searching the shortest path, obtaining a minimum energy diagram of the first key point of the queue, starting from the ending point, searching along the direction of minimum energy value, searching the next point in the queue, taking the point as the ending point, and cycling the process. The search path is the shortest path between the first and last points of the queue, i.e., the centerline of the coronary artery. The specific calculation method is shown in Table 2.
TABLE 2
In the step (3), the coronary topological graph is composed of nodes, edges and node characteristics, and the importance degree of surrounding nodes to a central node is measured in a mode of aggregating the node characteristics according to the edges in the follow-up process, so that the classification of the nodes is completed. Branches of the coronary are abstracted to nodes, and the interconnected branch nodes are connected by edges, whereas no branch nodes are connected by edges. For XRA images, the chapter shows images from four perspectives, based on clinical experience, to view and identify specific coronary branches, respectively.
Since the points of the coronary artery centerline are not distinguished from the intersection points when manually selecting the coronary artery centerline key points, the subsequent truncated centerline requires the point coordinates of the endpoint-located centerline points. Therefore, it is necessary to distinguish between an end point of 1 and an intersection point of 2, 3. The coronary endpoint is first found among all manually extracted keypoints. The principle of locating the coronary endpoints is that other centerline points within the neighborhood of the endpoint are all approximately distributed in a particular direction. Specifically, the two-by-two included angles of opposite endpoints of all center line points in the neighborhood are smaller, the included angles of every two adjacent points and the center point are calculated by traversing the center line points and the neighborhood center line points, and when all the included angles are smaller than a certain threshold value, the center point is the coronary endpoint.
Since a fast relay travel algorithm is used to search for the coronary centerline, the centerlines between the coronary endpoints are all relatively continuous, and the intersection point D, which is the relay point, is also in this centerline. Therefore, after the coronary artery end points are acquired, the intersection points with the degree of 3 are acquired first, the end points are paired two by two, and the center line point between the end points and the manually extracted intersection points are found. Any two centerlines AB, AC are then compared to each other, and if some centerline points overlap or are very similar, it can be inferred that the two centerlines of the overlapping portion represent the same coronary segment, and a group can be eliminated. And then recording the coordinates of a point D at the separated position of the two sections of central lines, matching with all the cross points, if the matching is successful, directly storing the connection relation of the three sections of central lines into the topological graph, and if the matching is unsuccessful, storing the point as the cross point and then storing the connection relation into the topological graph. And finally, considering the intersection points with the redundancy of 2, and storing the coronary artery segments and the connection relations at the two sides of the intersection points into the topological graph.
For the subsequent extraction of node features, the centerlines of the respective coronary segments are resampled so that the number of points of the centerlines is the same. The specific implementation steps of extracting the coronary topology by traversing the removal method are shown in table 3.
TABLE 3 Table 3
Preferably, in the step (4), vectors are usedTo represent each extracted centerline, where M represents the truncated coronary branch number, N represents the number of centerline points after coronary branch resampling, d represents the coordinate dimension, 3 in the CTA image and 2 in the XRA image;
the relationship between the different coronary branches is evaluated by calculating the relative position and growth direction of each coronary branch separately:
wherein,representing the relative distance features in the internal azimuthal vector field,/>representing the relative directional characteristics, +.>Representing the growth direction characteristics, for each coronary branch m, willAs its extracted centreline point, will +.>As the central line points of the other coronary branches except m, wherein i represents the ith point in the central line of the coronary branch, j represents the coordinate dimension, and also the start, mid and end respectively represent the starting point, the middle point and the end point in the central line, and I represents the splicing operation;
after calculating the relative distance, direction and growth direction characteristics of all branches of the coronary artery, cascading the characteristic information together to complete the construction of the internal azimuth vector field of the coronary artery, and providing node characteristics for the coronary artery topological graph
Wherein,the internal azimuthal vector field representing the coronary is the sum of three characteristic dimensions.
Preferably, in the step (5), 20 cases of complete branches in the 3D coronary artery data set are selected, the average of the directions and the lengths of the 3D coronary artery branches is calculated, a 3D coronary artery branch average model is constructed, the general characteristics of the 3D coronary artery are represented to the maximum extent, and the influence of individual variability on the average model is reduced; and calculating the horizontal angle difference alpha and the height angle difference beta of the main 3D average model main and each 2D coronary artery main, forming angle guiding characteristics, and finishing rough guiding.
Preferably, in the step (5), the features are calculated separately for each view, and the 3D main vector used for each view is different. LAD branches in 3D coronary artery are used in left crown left front oblique position, LCX branches are used in left crown right front oblique position, RCA branches are used in right crown left front oblique position, R-PDB branches are used in right crown right front oblique position, main supporting vectors of 2D coronary artery are determined by first key point and last key point extracted manually, two main supporting vectors are normalized to unit vectors at the same time, starting points are assumed to be located at the original point of the same coordinate system, then the first two items of 3D vectors and 2D vectors are taken to calculate horizontal angle difference, then the third item of 3D vectors and the first two items of 3D vectors are taken to calculate height angle difference, radian is used to represent the height angle difference, and angle guiding characteristics are obtained after the horizontal angle difference and the height angle difference are spliced
Wherein,unit vector representing the first two components of the 3D main vector,/->Representing the 2D main support unit vector,>a unit vector representing the last term of the 3D main vector, II represents modulo calculation, cos -1 Representing the inverse function of cos (x), sin -1 Representing the inverse of sin (x);
encoding the results of the 3D coronary branches, for a total of 9 branches, then the encoding range is [1,9];
then aligning the 2D main vector complement 0 with the 3D main vector, and calculating the angle difference theta between the two vectors; projecting length and radius information of the 3D main branch into each 2D branch vector by an angle difference θ
a＝Rbcosθ+c (4.4)
Wherein,a unit vector representing a 3D main vector, theta is an included angle between the 2D vector and the 3D vector, a is a fusion characteristic, b is length and radius information of a main branch in a 3D azimuth vector field, c is length and radius information of each branch in the 2D azimuth vector field, R is identification result code of the 3D coronary main branch, and the range is [1,9]]Representing the branches, respectively, each view will choose a different code.
Preferably, in said step (6), it is assumed that N represents all coronary branches, for each coronary branchAll belonging to one of three types of topology, assumed to be U t T.epsilon.1, 2,3, node layer will calculate +.>In type U t Neighborhood node->Weight of +.>Represents->Neighborhood branch pair->Importance of (2)
Wherein the intent is a layer of feedforward neural network, using the LeakyReLU activation function, f ni Representative ofNode characteristics of f tj Represents U t Neighborhood nodes of type->Is only when +.>Neighborhood node of (c) belonging to U t Calculating the attribute when the type is found, otherwise, the attribute=0; then obtaining the weight coefficient by normalization>
The branches in each type of topological graph are weighted and aggregated into a joint graph node U through weight coefficients nt Calculating an average result obtained after multiple weighted aggregation by using a multi-head attention mechanism
Wherein S represents the number of independent calculations of the multi-headed attention,representing the weight coefficient in the s-th calculation, σ represents the LeakyReLU activation function, and in order to reduce the dimension of the output matrix, the final output vector U is calculated by an averaging operation nt ，U nt A joint graph node of t type; to obtain enough learning ability to calculate weights between joint graph nodes, a learnable weight matrix W is used to assist in extracting the influence weights between joint graph nodes
Wherein,representing the weight between the ith joint graph node, N nj Representing a j-th neighborhood joint graph node, wherein the unit represents a layer of feedforward neural network, and the activation function is LeakyReLU;
normalizing the weight of each joint graph node through a softmax function to obtain a weight coefficient, and finally expressing the fused coefficient asFinally, an average output vector eta is obtained through a multi-head attention mechanism ni
Wherein U is ni Representing neighborhood joint graph nodes.
It will be understood by those skilled in the art that all or part of the steps in implementing the above embodiment method may be implemented by a program to instruct related hardware, where the program may be stored in a computer readable storage medium, where the program when executed includes the steps of the above embodiment method, and the storage medium may be: ROM/RAM, magnetic disks, optical disks, memory cards, etc. Accordingly, the present invention also includes a coronary branch identification device of an XRA image, corresponding to the method of the present invention. The device comprises:
the coronary artery adaptive parameter rapid pre-segmentation module is configured to use an adaptive framework nnUnet based on 2D U-Net, adopts depth separable convolution to replace traditional convolution, reduces the parameter quantity under the condition of unchanged segmentation precision, and quickens the reasoning speed; in preprocessing, in order to reduce the amount of computation, the framework uses clipping preprocessing for the zero-value regions;
the rapid relay travel optimization method draws a coronary artery segment path module configured to extract a coronary artery centerline using a semi-automatic approach based on travel optimization;
constructing a multi-mode coronary artery topological graph module by using a traversal elimination method, wherein the module is configured to take the traversal elimination method as a core, finish automatic identification of a coronary artery central line endpoint, identify branch boundaries and resample central line points to enable the number of the branch boundaries to be consistent, construct a coronary artery topological graph, intercept the consistent coronary artery central line points into identifiable branches, and acquire nodes and edges required in the topological graph;
the coronary segment internal azimuth vector field feature extraction module is configured to define an internal azimuth vector field to simulate the dependence of different coronary branches on other internal branches, evaluate the relation among different coronary branches by respectively calculating the relative position and the growth direction of each coronary branch, and after calculating the relative distance, the direction and the growth direction features of all the branches of the coronary, cascade the feature information together to complete the construction of the coronary internal azimuth vector field and provide node features for the coronary topological graph;
the angle guiding and proportion fusing module of the three-dimensional coronary artery features is configured to divide the XRA image into four visual angles to respectively observe different branches, guide the 2D features by using the 3D coronary artery branch recognition result, extract the angle guiding features, then fuse the 3D features and the 2D features to obtain fused features, and fully supplement the missing space information;
the coronary feature fusion module is configured to fuse three types of coronary features together through a joint graph attention network structure, improve feature expression capacity, the joint graph attention network is composed of two layers of graph attention, namely a node layer and a joint layer, the weight coefficient of a progressive automatic learning neighborhood node to a central node is obtained through weighted aggregation, the node layer learns the weight of the neighborhood node in each group of coronary topological graph respectively, the joint graph nodes are calculated through aggregation, the three joint graph nodes respectively represent different types of coronary feature information, the joint layer extracts the correlation of the three joint graph nodes, the importance degree of the three branch information to the branch class is measured, and the branch identification result is finally weighted aggregated.
The present invention is not limited to the preferred embodiments, but can be modified in any way according to the technical principles of the present invention, and all such modifications, equivalent variations and modifications are included in the scope of the present invention.
Claims (10)
- A method of coronary branch identification for xra images, characterized by: which comprises the following steps:(1) The coronary artery adaptive parameter is rapidly pre-segmented, an adaptive framework nnUnet based on 2D U-Net is used, depth separable convolution is adopted to replace traditional convolution, the parameter quantity is reduced under the condition that the segmentation precision is unchanged, and the reasoning speed is increased; in preprocessing, in order to reduce the amount of computation, the framework uses clipping preprocessing for the zero-value regions;(2) Drawing a coronary artery segment path by a rapid relay travel optimization method, and extracting a coronary artery central line by a semi-automatic method based on travel optimization;(3) Constructing a multi-mode coronary artery topological graph by using a traversal elimination method as a core, completing automatic identification of a coronary artery central line endpoint, identifying branch boundaries and resampling central line points to enable the number of the branch boundaries to be consistent, constructing the coronary artery topological graph, cutting the consistent coronary artery central line points into identifiable branches, and obtaining nodes and edges required in the topological graph;(4) Extracting features of internal azimuth vector fields of coronary artery segments, defining internal azimuth vector fields to simulate the dependence of different coronary artery branches on other internal branches, evaluating the relation between different coronary artery branches by respectively calculating the relative position and growth direction of each coronary artery branch, cascading the feature information together to complete the construction of the internal azimuth vector fields of the coronary artery after calculating the features of the relative distance, direction and growth direction of all the branches of the coronary artery, and providing node features for a coronary artery topological graph;(5) Angle guiding and proportion fusion of three-dimensional coronary artery characteristics are carried out, an XRA image is divided into four visual angles to respectively observe different branches, 3D coronary artery branch recognition results are utilized to guide 2D characteristics, angle guiding characteristics are extracted, then 3D characteristics and 2D characteristics are fused to obtain fusion characteristics, and missing space information is fully supplemented;(6) The three types of coronary artery characteristics are fused together through a joint graph attention network structure, the characteristic expression capacity is improved, the joint graph attention network consists of two layers of graph attention, namely a node layer and a joint layer, the weight coefficient of a progressive automatic learning neighborhood node to a central node is obtained through weighted aggregation, the node layer learns the weight of the neighborhood node in each group of coronary artery topological graph respectively, the joint graph nodes are calculated through aggregation, the three joint graph nodes respectively represent different types of coronary artery characteristic information, the joint layer extracts the correlation relation of the three joint graph nodes, the importance degree of the three branch information to the branch class is measured, and the branch identification result is finally weighted aggregated.
- 2. The method of coronary branch identification of an XRA image of claim 1, wherein: in the step (1) described above, the step of (c) is performed,in terms of the network, the U-Net network architecture includes an encoder, a hop junction, a decoder, an encoder consisting of a depth separable convolutional layer, a leak ReLU activation function, instance Normalization,2 x 2 max pooling layer, U-Net extracting low-level and high-level semantic feature map of coronary image, fusing semantic features of different scales; the convolution layer adopts depth separable convolution, the convolution calculation is divided into two steps, the number of convolution kernels is the same as that of the channels of the previous layer, the channels and the convolution kernels are in one-to-one correspondence, the size of the convolution kernels is 1 multiplied by M, M is the number of the channels of the previous layer, and the point-to-point convolution operation carries out weighted combination on the feature map of the first step in the depth direction to generate a new feature map;in the aspect of network training, the training super-parameters are adaptively adjusted according to the size of the coronary artery image, and the frame nnU-Net monitors the occupation condition of the video memory in real time and automatically sets the batch size and the image block size to prevent the explosion of the video memory.
- 3. The method of coronary branch identification of an XRA image of claim 2, wherein: in the step (1), the Batch size is adjusted in preference to the Batch; the size of Patch ensures that at least one third of the sampled area is coronary; using the sum of the cross entropy loss and the Dice loss as a loss function; the learning rate is adaptively adjusted according to the exponential average of the loss function.
- 4. The method of coronary branch identification of an XRA image of claim 3, wherein: in the step (2), the branch end points and the bifurcation points of the coronary artery of the concerned part are manually determined as key points, and the shortest path searching and drawing among a plurality of key points is realized by using an improved advancing optimization method; constructing an energy map by utilizing the image characteristic difference of the coronary blood vessel and the background, wherein in the energy map, the energy value at the center of the coronary blood vessel is smaller, and the energy value towards the edge of the blood vessel and the background direction is gradually increased; manually setting a starting point, solving an Eikonal equation on a speed map by using a rapid advancing method to obtain a minimum energy map corresponding to the point, setting a termination point, reversely searching the starting point to enable the energy of the path to be minimum, and obtaining a shortest path which is used as the central line of the coronary artery.
- 5. The method of coronary branch identification of an XRA image of claim 4, wherein: in the step (2), the step of (c),regarding the solving of the energy map, first a coronary image I is defined in which the pixel values of the centerline position have smaller values, assuming two points p on the map 1 、p 2 ，B p Is p 1 、p 2 A set of paths between, defining an energy map function E:the expression is as follows:E(δ)＝∫I(δ(z))fz (2.1)wherein δ is a path between two points, z is a length parameter such that δ, where the energy E (δ) is the smallest, is the coronary centerline between the two points; when searching the center line, find the corresponding p 1 Is defined by the minimum energy diagram phi:any point on the image +.>The value in the minimum energy diagram is p 1 To->Energy value corresponding to coronary centerline:satisfies the Eikonal equation:the fast travelling method is a numerical solution for Eikonal equation, a stable solution for Eikonal equation is solved by using a reverse difference method, a first-order forward and backward difference operator is used for replacing a differential quotient, and the calculation of a single pixel point is as follows:wherein h is x And h y Representing the pixel difference size in the x, y direction of the image.
- 6. The method of coronary branch identification of an XRA image of claim 5, wherein: in the step (4), vectors are usedRepresenting each extracted centerline, where M represents a truncated coronary branchThe number, N, represents the number of centerline points after coronary branch resampling, d represents the coordinate dimension, 3 in CTA image and 2 in XRA image;the relationship between the different coronary branches is evaluated by calculating the relative position and growth direction of each coronary branch separately:wherein,representing the relative distance features in the internal azimuthal vector field, a->Representing the relative directional characteristics, +.>Representing the growth direction characteristics, for each coronary branch m, willAs its extracted centreline point, will +.>As the centerline points of the coronary branches except m, i represents the ith point in the centerline of the coronary branch, j represents the coordinate dimension, and also the start, mid and end respectively represent the starting point and the middle point in the centerlineEnd point, | represents the splicing operation;after calculating the relative distance, direction and growth direction characteristics of all branches of the coronary artery, cascading the characteristic information together to complete the construction of the internal azimuth vector field of the coronary artery, and providing node characteristics for the coronary artery topological graphWherein,the internal azimuthal vector field representing the coronary is the sum of three characteristic dimensions.
- 7. The method of coronary branch identification of an XRA image of claim 6, wherein: in the step (5), 20 complete branches in the 3D coronary artery data set are selected, the average of the directions and the lengths of the 3D coronary artery branches is calculated, a 3D coronary artery branch average model is constructed, the general characteristics of the 3D coronary artery are represented to the maximum extent, and the influence of individual variability on the average model is reduced; and calculating the horizontal angle difference alpha and the height angle difference beta of the main 3D average model main and each 2D coronary artery main, forming angle guiding characteristics, and finishing rough guiding.
- 8. The method of coronary branch identification of an XRA image of claim 7, wherein: in the step (5), the features are calculated for each view angle respectively, the 3D main vector used for each view angle is also different, the left crown left front oblique position uses the LAD branch in the 3D coronary artery, the left crown right front oblique position uses the LCX branch, the right crown left front oblique position uses the RCA branch, the right crown right front oblique position uses the R-PDB branch, the main vector of the 2D coronary artery is determined by the first key point and the last key point extracted manually, the two main vectors are normalized to the unit vector at the same time, the starting points are assumed to be at the origin of the same coordinate system, then the first two items of the 3D vector and the 2D vector are taken to calculate the horizontal angle difference, the third item of the 3D vector and the first two items of the 3D vector are taken to calculate the height angle difference, the horizontal angle difference and the height angle difference are all expressed by radians, and the angle guiding feature is obtained after the horizontal angle difference and the height angle difference are splicedWherein,unit vector representing the first two components of the 3D main vector,/->Representing the 2D main support unit vector,>a unit vector representing the last term of the 3D main vector, II represents modulo calculation, cos -1 Representing the inverse function of cos (x), sin -1 Representing the inverse of sin (x);encoding the results of the 3D coronary branches, for a total of 9 branches, then the encoding range is [1,9]; then aligning the 2D main vector complement 0 with the 3D main vector, and calculating the angle difference theta between the two vectors; projecting length and radius information of the 3D main branch into each 2D branch vector by an angle difference θa＝Rb cosθ+c (4.4)Wherein,unit vector, θ, representing the 3D principal vectorFor the included angle between the 2D vector and the 3D vector, a is the fusion characteristic, b is the length and radius information of the main branch in the 3D azimuth vector field, c is the length and radius information of each branch in the 2D azimuth vector field, R is the identification result code of the 3D coronary main branch, and the range is [1,9]]Representing the branches, respectively, each view will choose a different code.
- 9. The method of coronary branch identification of an XRA image of claim 8, wherein: in said step (6), it is assumed that N represents all coronary branches, for each coronary branchAll belonging to one of three types of topology, assumed to be U t T.epsilon.1, 2,3, node layer will calculate +.>In type U t Neighborhood node->Weight of +.>Represents->Neighborhood branch pair->Importance of (2)Wherein the intent is a layer of feedforward neural network, using the LeakyReLU activation function, f ni Representative ofNode characteristics of f tj Represents U t Neighborhood nodes of type->Is only when +.>Neighborhood node of (c) belonging to U t Calculating the attribute when the type is found, otherwise, the attribute=0; then obtaining the weight coefficient by normalization>The branches in each type of topological graph are weighted and aggregated into a joint graph node U through weight coefficients nt Calculating an average result obtained after multiple weighted aggregation by using a multi-head attention mechanismWherein S represents the number of independent calculations of the multi-headed attention,representing the weight coefficient in the s-th calculation, σ represents the LeakyReLU activation function, and in order to reduce the dimension of the output matrix, the final output vector U is calculated by an averaging operation nt ，U nt A joint graph node of t type; to obtain enough learning ability to calculate weights between joint graph nodes, a learnable weight matrix W is used to assist in extracting the influence weights between joint graph nodesWherein,representing the weight between the ith joint graph node, N nj Representing a j-th neighborhood joint graph node, wherein the unit represents a layer of feedforward neural network, and the activation function is LeakyReLU; normalizing the weights of each joint graph node by a softmax function to obtain weight coefficients, and finally expressing the fused coefficients as +.>Finally, an average output vector eta is obtained through a multi-head attention mechanism niWherein U is ni Representing neighborhood joint graph nodes.
- 10. The apparatus of the coronary branch identification method of an XRA image of claim 1, wherein: it comprises the following steps:the coronary artery adaptive parameter rapid pre-segmentation module is configured to use an adaptive framework nnUnet based on 2D U-Net, adopts depth separable convolution to replace traditional convolution, reduces the parameter quantity under the condition of unchanged segmentation precision, and quickens the reasoning speed; in preprocessing, in order to reduce the amount of computation, the framework uses clipping preprocessing for the zero-value regions;the rapid relay travel optimization method draws a coronary artery segment path module configured to extract a coronary artery centerline using a semi-automatic approach based on travel optimization;constructing a multi-mode coronary artery topological graph module by using a traversal elimination method, wherein the module is configured to take the traversal elimination method as a core, finish automatic identification of a coronary artery central line endpoint, identify branch boundaries and resample central line points to enable the number of the branch boundaries to be consistent, construct a coronary artery topological graph, intercept the consistent coronary artery central line points into identifiable branches, and acquire nodes and edges required in the topological graph;the coronary segment internal azimuth vector field feature extraction module is configured to define an internal azimuth vector field to simulate the dependence of different coronary branches on other internal branches, evaluate the relation among different coronary branches by respectively calculating the relative position and the growth direction of each coronary branch, and after calculating the relative distance, the direction and the growth direction features of all the branches of the coronary, cascade the feature information together to complete the construction of the coronary internal azimuth vector field and provide node features for the coronary topological graph;the angle guiding and proportion fusing module of the three-dimensional coronary artery features is configured to divide the XRA image into four visual angles to respectively observe different branches, guide the 2D features by using the 3D coronary artery branch recognition result, extract the angle guiding features, then fuse the 3D features and the 2D features to obtain fused features, and fully supplement the missing space information;the coronary feature fusion module is configured to fuse three types of coronary features together through a joint graph attention network structure, improve feature expression capacity, the joint graph attention network is composed of two layers of graph attention, namely a node layer and a joint layer, the weight coefficient of a progressive automatic learning neighborhood node to a central node is obtained through weighted aggregation, the node layer learns the weight of the neighborhood node in each group of coronary topological graph respectively, the joint graph nodes are calculated through aggregation, the three joint graph nodes respectively represent different types of coronary feature information, the joint layer extracts the correlation of the three joint graph nodes, the importance degree of the three branch information to the branch class is measured, and the branch identification result is finally weighted aggregated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310602873.9A CN117011619B (en) | 2023-05-26 | Method and apparatus for coronary artery branch identification in XRA images |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310602873.9A CN117011619B (en) | 2023-05-26 | Method and apparatus for coronary artery branch identification in XRA images |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117011619A true CN117011619A (en) | 2023-11-07 |
| CN117011619B CN117011619B (en) | 2025-12-26 |
Family
ID=
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119991617A (en) * | 2025-01-20 | 2025-05-13 | 北京理工大学 | Method, device, equipment and medium for identifying blood vessel branches in angiography images |
| US20250218053A1 (en) * | 2023-12-12 | 2025-07-03 | AtomBeam Technologies Inc. | System and methods for multimodal series transformation for optimal compressibility with neural upsampling |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111667456A (en) * | 2020-04-28 | 2020-09-15 | 北京理工大学 | Method and device for detecting vascular stenosis in coronary artery X-ray sequence radiography |
| CN111951277A (en) * | 2020-07-28 | 2020-11-17 | 杭州电子科技大学 | A Coronary Artery Segmentation Method Based on CTA Image |
| WO2021081771A1 (en) * | 2019-10-29 | 2021-05-06 | 未艾医疗技术(深圳)有限公司 | Vrds ai medical image-based analysis method for heart coronary artery, and related devices |
| CN112785551A (en) * | 2020-12-30 | 2021-05-11 | 杭州电子科技大学 | Coronary artery segmentation method based on deep learning |
| CN113935889A (en) * | 2021-09-08 | 2022-01-14 | 顾力栩 | Method, system and medium for automatic 2D/3D coronary artery registration |
| CN114627128A (en) * | 2022-02-11 | 2022-06-14 | 珠海横乐医学科技有限公司 | Coronary artery XRA segmentation method and device based on residual attention network and readable medium |
| CN115063384A (en) * | 2022-06-29 | 2022-09-16 | 北京理工大学 | SP-CTA image coronary artery segmentation method and device based on feature alignment domain |
| CN115713626A (en) * | 2022-11-21 | 2023-02-24 | 山东省人工智能研究院 | 3D coronary artery CTA plaque identification method based on deep learning |
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021081771A1 (en) * | 2019-10-29 | 2021-05-06 | 未艾医疗技术(深圳)有限公司 | Vrds ai medical image-based analysis method for heart coronary artery, and related devices |
| CN111667456A (en) * | 2020-04-28 | 2020-09-15 | 北京理工大学 | Method and device for detecting vascular stenosis in coronary artery X-ray sequence radiography |
| CN111951277A (en) * | 2020-07-28 | 2020-11-17 | 杭州电子科技大学 | A Coronary Artery Segmentation Method Based on CTA Image |
| CN112785551A (en) * | 2020-12-30 | 2021-05-11 | 杭州电子科技大学 | Coronary artery segmentation method based on deep learning |
| CN113935889A (en) * | 2021-09-08 | 2022-01-14 | 顾力栩 | Method, system and medium for automatic 2D/3D coronary artery registration |
| CN114627128A (en) * | 2022-02-11 | 2022-06-14 | 珠海横乐医学科技有限公司 | Coronary artery XRA segmentation method and device based on residual attention network and readable medium |
| CN115063384A (en) * | 2022-06-29 | 2022-09-16 | 北京理工大学 | SP-CTA image coronary artery segmentation method and device based on feature alignment domain |
| CN115713626A (en) * | 2022-11-21 | 2023-02-24 | 山东省人工智能研究院 | 3D coronary artery CTA plaque identification method based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 水浩淼;李勤;刘越;王涌天;: "冠状动脉三维重建前期工作的关键技术研究", 中国医学影像技术, no. 03, 20 March 2006 (2006-03-20) * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20250218053A1 (en) * | 2023-12-12 | 2025-07-03 | AtomBeam Technologies Inc. | System and methods for multimodal series transformation for optimal compressibility with neural upsampling |
| US12437448B2 (en) * | 2023-12-12 | 2025-10-07 | AtomBeam Technologies Inc. | System and methods for multimodal series transformation for optimal compressibility with neural upsampling |
| CN119991617A (en) * | 2025-01-20 | 2025-05-13 | 北京理工大学 | Method, device, equipment and medium for identifying blood vessel branches in angiography images |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115457395B (en) | Lightweight remote sensing target detection method based on channel attention and multi-scale feature fusion | |
| Du et al. | A general pipeline for 3d detection of vehicles | |
| CN114119975B (en) | Cross-modal instance segmentation method guided by language | |
| Du et al. | Car detection for autonomous vehicle: LIDAR and vision fusion approach through deep learning framework | |
| WO2024060395A1 (en) | Deep learning-based high-precision point cloud completion method and apparatus | |
| CN110060263A (en) | Medical image segmentation method, segmentation device, segmentation system and computer readable medium | |
| WO2022100379A1 (en) | Object attitude estimation method and system based on image and three-dimensional model, and medium | |
| CN114187310A (en) | Large-scale point cloud segmentation method based on octree and PointNet ++ network | |
| CN114792372A (en) | A method and system for semantic segmentation of 3D point cloud based on multi-head and two-level attention | |
| CN112686902B (en) | A Two-Stage Computational Method for the Identification and Segmentation of Midbrain Gliomas in Magnetic Resonance Imaging | |
| CN111127538A (en) | A 3D reconstruction method for multi-view images based on convolutional cyclic encoding-decoding structure | |
| CN109359534B (en) | A method and system for extracting geometric features of three-dimensional objects | |
| CN111968138A (en) | Medical image segmentation method based on 3D dynamic edge insensitivity loss function | |
| KR102305230B1 (en) | Method and device for improving accuracy of boundary information from image | |
| Choi et al. | mbest: Realtime deformable linear object detection through minimal bending energy skeleton pixel traversals | |
| US12293555B2 (en) | Method and device of inputting annotation of object boundary information | |
| Casaca et al. | Laplacian coordinates: Theory and methods for seeded image segmentation | |
| CN120107601B (en) | Three-dimensional hepatobiliary duct, hepatic artery, portal vein and hepatic vein image segmentation method and system | |
| CN117784815B (en) | Path planning and defect identification method and device for unmanned aerial vehicle inspection | |
| CN115937249A (en) | A target tracking method and device based on multi-prediction output alignment of Siamese network | |
| CN116109649A (en) | 3D point cloud instance segmentation method based on semantic error correction | |
| CN113592015A (en) | Method and device for positioning and training feature matching network | |
| CN114022458A (en) | Skeleton detection method, apparatus, electronic device and computer-readable storage medium | |
| Gonzalez-Huitron et al. | Jaccard distance as similarity measure for disparity map estimation | |
| CN116363473A (en) | A Lunar Impact Crater Identification Method Based on Data Fusion and Transfer Learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |