A myocardial bridge detection apparatus in a coronary angiography sequence

Abstract

The application relates to a myocardial bridge detection device in a coronary angiography sequence, comprising: a coronary vessel segmentation module for segmenting a coronary vessel through a neural network model, the model being composed of a CNN and a transformer structure, and being capable of effectively extracting local information and global information of the vessel; a coronary vessel information calculation module for calculating information including the diameter, length and start-stop coordinates of the vessel according to a vessel segmentation map, establishing a coronary vessel tree containing all the vessels, and calculating matching vessels between the angiography sequences; and a coronary myocardial bridge detection module for realizing the detection of the coronary myocardial bridge by calculating the stenosis degree of the same vessel in the angiography sequence.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and more particularly to an automatic detection method for myocardial bridging in coronary angiography sequences. Background Technology

[0002] Coronary artery myocardial bridging is a congenital coronary artery malformation where a segment of the coronary artery, which should run along the epicardium, penetrates into the myocardium, and the myocardium covering it is called a myocardial bridge. Myocardial bridging often occurs in the left anterior descending artery. During myocardial contraction, it compresses the coronary artery, leading to insufficient blood perfusion to the corresponding part of the myocardium during systole, causing symptoms of myocardial ischemia such as angina pectoris, arrhythmia, and myocardial infarction.

[0003] Clinically, coronary angiography is the gold standard for diagnosing coronary artery disease, especially myocardial bridging. When diagnosing a patient's condition, physicians need to quickly observe vascular imaging and obtain diagnostic results rapidly. This requires physicians to have extensive clinical experience and knowledge of anatomical physiology. However, subjectivity inevitably influences the diagnostic results. In the field of computer-aided diagnosis, to achieve automated diagnosis of myocardial bridging, it is necessary to accurately identify vessels from the angiographic sequence and calculate the degree of vascular stenosis. However, due to the complex background structure and uneven imaging intensity of coronary angiography images, achieving automated detection of myocardial bridging during coronary angiography presents certain challenges, but also holds significant importance.

[0004] Currently, there are few methods for detecting myocardial bridging. Patent CN114972242B discloses a method for detecting myocardial bridging based on cardiac CT angiography (CTA). This method uses deep learning to predict the probability of blood vessels being covered by myocardium to detect myocardial bridging. However, myocardial bridging exhibits periodic contraction and narrowing, while CTA images the state of the heart at a specific moment, making direct detection of myocardial bridging in CTA extremely difficult. Summary of the Invention

[0005] To overcome the above shortcomings, this invention provides a device for detecting myocardial bridging in coronary angiography sequences, enabling relatively accurate detection of myocardial bridging during coronary angiography. The technical solution is as follows:

[0006] A device for detecting myocardial bridging in a coronary angiography sequence, comprising:

[0007] A coronary artery segmentation module is used to segment coronary arteries using a neural network model. This model consists of CNN and transformer structures, and can effectively extract local and global information about the blood vessels. The neural network model includes:

[0008] (1.1) The neural network model is an end-to-end model. The encoding module consists of a residual module, an improved transformer encoding module, and a max pooling layer. The decoding module consists of a residual module, an improved transformer decoding module, a deconvolution layer, and skip connections.

[0009] The improved transformer codec structure formula is as follows:

[0010]

[0011] In the formula, K and V are obtained by expanding a small-sized feature map smaller than the original feature map through two different 1x1 convolution and downsampling operations;

[0012] (1.2) The coding module consists of a 5-layer structure, with each layer connected by a max pooling layer. The first layer consists of 2 residual modules, and the other layers consist of 1 residual module and 1 improved transformer coding module.

[0013] (1.3) The decoding module also consists of a 5-layer structure. Except for the 5th layer, each of the other layers has a residual module. The 2nd, 3rd and 4th layers achieve feature map upsampling and feature fusion of the corresponding layer in the encoding module through the improved transformer decoding structure. The 1st and 2nd layers are connected by a deconvolution layer. At the same time, the 1st layer achieves feature fusion with the 1st layer in the encoding module through a skip connection.

[0014] The coronary artery information calculation module is used to calculate information including the diameter, length, and origin-end coordinates of blood vessels based on the vessel segmentation map, build a coronary artery tree containing all blood vessels, and calculate matching vessels between angiography sequences; including:

[0015] (2.1) Extract the vascular skeleton from the vascular segmentation map;

[0016] (2.2) Extract the endpoints and intersections in the vascular skeleton, separate the blood vessels to obtain vascular segments, and remove the spiky blood vessels;

[0017] (2.3) Record the pixel coordinates at both ends of the blood vessel segment as the starting and ending coordinates of the blood vessel, calculate the number of pixels in the blood vessel segment as the blood vessel length, fit the blood vessel straight line using the least squares method, then obtain the blood vessel normal, and then calculate the distance between the intersection of the blood vessel normal and the blood vessel edge as the diameter of the blood vessel.

[0018] (2.4) Starting with the angiography catheter, establish a trigeminal vascular tree based on spatial location;

[0019] (2.5) Matching of blood vessels in coronary angiography sequences based on information about the vessel segments;

[0020] The coronary myocardial bridging detection module is used to detect coronary myocardial bridging by calculating the degree of stenosis of the same vessel in the angiography sequence.

[0021] Further, in (2.1), the Zhang-Suen method is used to extract the vascular skeleton from the vascular segmentation map.

[0022] Furthermore, the method for extracting the vascular skeleton intersection point in (2.2) is as follows: for a certain pixel point, if the sum of its 8 neighboring pixels is ≥3 and the sum of the pixels of its three adjacent neighboring pixels is ≤1, then this pixel point is judged as a vascular skeleton intersection point.

[0023] Furthermore, the spatial location-based tritree vascular tree (2.4) includes:

[0024] (2.4.1) Starting from the angiography catheter, determine the sub-nodes and their number for each vascular node sequentially, and use the Manhattan distance d between the coordinates of the node's endpoint and the coordinates of its sub-node's starting point. n To determine the child nodes of a node, the coordinates of the child node whose origin is less distant from the coordinates of the node's endpoints are the child node's origin coordinates. n Less than or equal to threshold T n When the node is in the child node position, it is a child node of the node; otherwise, it is not a child node.

[0025] (2.4.2) The position of a child node in the vascular tree is determined based on the spatial positions of multiple child nodes. When a node has two child nodes, the endpoint coordinates p1(x1,y1) of the node and the starting coordinates p2(x2,y2) and p3(x3,y3) of the two child nodes form two vectors p1p2 and p2p3. The cross product of the two vectors is given by the formula:

[0026] V=(x2-x1)×(y3-y2)-(y2-y1)×(x3-x2)

[0027] Here, V is the cross product of vectors. When V is greater than 0, p1, p2, and p3 are arranged in counterclockwise order, so p2 is the first child node, and vice versa.

[0028] When there are 3 child nodes, the coordinates of the endpoint p1 of the node and the coordinates of the starting points p2, p3, and p4 of the 3 child nodes form three pairs of vectors p1p2 and p2p3, p1p2 and p2p4, and p1p3 and p3p4. Calculate the cross product of the three pairs of vectors respectively, and determine the order of the 4 points based on the value of the cross product.

[0029] (2.4.3) Based on the growth characteristics of coronary angiography, assign an ID value to each vessel from top to bottom.

[0030] Furthermore, the matching of vessels in the coronary angiography sequence in (2.5) includes:

[0031] (2.5.1) For blood vessel A1 in the previous frame and blood vessel A2 in the current frame, the starting coordinates and ending coordinates of the two blood vessels are S1(x) and S2(x) respectively. s1 ,y s1 ) and S2(x s2 ,y s2 ), E1(x e1 ,y e1 ) and E2(x e2 ,y e2 ), calculate the coordinates S1(x) of the vessel origin in the preceding and following angiographic images. s1 ,y s1 ) and S2(x s2 ,y s2 The endpoint coordinates are E1(x). e1 ,y e1 ) and E2(x e2 ,y e2 Manhattan distance d s and d e ;

[0032] (2.5.2) Let the lengths of blood vessels A1 and A2 be L1 and L2, respectively. That is, calculate the ratio of the lengths of blood vessels in the preceding and following angiographic images:

[0033]

[0034] (2.5.3) Determine the distance threshold T and the proportion threshold R, and match blood vessels according to the conditional formula, which is:

[0035]

[0036] Furthermore, the coronary myocardial bridging detection module includes:

[0037] (3.1) Calculate the degree of vascular stenosis using the following formula:

[0038]

[0039] In the formula, P represents the degree of vascular stenosis, and D... min D represents the minimum diameter of a blood vessel. f and D b Represents D min Normal diameter values ​​for both sides of the blood vessels; when the blood vessels are peripheral vessels, D min Represents the local minimum value of a blood vessel;

[0040] (3.2) Based on the changes in the degree of vascular stenosis between consecutive frames, when the change in the degree of stenosis exceeds the threshold T s At that time, it was diagnosed as myocardial bridging;

[0041] (3.3) In cases of overall vascular stenosis, myocardial bridging is detected by changes in the average diameter. Attached Figure Description

[0042] Figure 1 Flowchart for coronary angiography and myocardial bridging detection;

[0043] Figure 2 Diagram of the vascular segmentation network structure in coronary angiography;

[0044] Figure 3 Coronary artery segmentation result image;

[0045] Figure 4 Visualization of the coronary artery tree;

[0046] Figure 5 Coronary artery matching results image;

[0047] Figure 6 Diagram of stenosis detection;

[0048] Figure 7 Image showing the results of coronary myocardial bridging examination. Detailed Implementation

[0049] To make the technical problem, technical solution, and advantages of this invention clearer, a detailed description will be provided below in conjunction with the accompanying drawings and specific embodiments. Figure 1 As shown, the device for detecting myocardial bridging in a coronary angiography sequence includes:

[0050] The coronary artery segmentation module is used to segment coronary arteries using a neural network model. This model consists of CNN and transformer structures, which can effectively extract local and global information about the blood vessels.

[0051] The coronary artery information calculation module is used to calculate information including the diameter, length, and origin and end coordinates of the vessels based on the vessel segmentation map, build a coronary artery tree containing all vessels, and calculate matching vessels between angiography sequences.

[0052] The coronary myocardial bridging detection module is used to detect coronary myocardial bridging by calculating the degree of stenosis of the same vessel in the angiography sequence.

[0053] The network structure of the coronary artery segmentation module of this invention is as follows: Figure 2 As shown, the input to the neural network is a 512x512 coronary angiography image, and the neural network model includes:

[0054] (1.1) The neural network model is an end-to-end model. The encoding module consists of a residual module, an improved transformer encoding module, and a max pooling layer. The decoding module consists of a residual module, an improved transformer decoding module, a deconvolution layer, and skip connections.

[0055] (1.2) The coding module consists of a 5-layer structure, with each layer connected by a max pooling layer. The first layer consists of 2 residual modules, and the other layers consist of 1 residual module and 1 improved transformer coding module.

[0056] (1.3) The decoding module also consists of a 5-layer structure. Except for the 5th layer, each of the other layers has a residual module. The 2nd, 3rd and 4th layers achieve feature map upsampling and feature fusion of the corresponding layer in the encoding module through the improved transformer decoding structure. The 1st and 2nd layers are connected by a deconvolution layer. At the same time, the 1st layer achieves feature fusion with the 1st layer in the encoding module through a skip connection.

[0057] The improved transformer's encoding and decoding structures differ in their inputs. The encoder's input is the output of the residual module, while the decoder's input consists of the output of the corresponding encoding layer and the upsampled output of the decoding module. The output of the corresponding encoding layer is obtained by performing a 1x1 convolution and unpacking to get Q. The upsampled output of the decoding module is obtained by unpacking it using two different 1x1 convolutions and downsampling operations. and

[0058] The improved transformer encoding / decoding structure formula in (1.1) is as follows:

[0059]

[0060] In the formula and It is obtained by convolving a 16x16 feature map with a 1x1 size and then unfolding it, which greatly reduces the amount of computation and is suitable for training and learning on small sample data. Figure 3 The image shown is the segmentation result of the coronary artery.

[0061] The coronary artery information calculation module of this invention includes:

[0062] (2.1) The vascular skeleton in the vascular segmentation map was extracted using the Zhang-Suen method;

[0063] (2.2) Extract the endpoints and intersections in the vascular skeleton, separate the blood vessels to obtain vascular segments, and remove erroneous blood vessels caused by unclear blood vessel edges and defects in the skeleton extraction algorithm. If the length of the blood vessel segment is less than L, it is judged as a erroneous blood vessel. The value of L is 6.

[0064] (2.3) Record the pixel coordinates of the two segments of the blood vessel as the starting and ending coordinates of the blood vessel, calculate the number of pixels of the blood vessel segment as the blood vessel length, fit the blood vessel straight line using the least squares method, and then obtain the blood vessel normal. Then calculate the distance between the intersection of the blood vessel normal and the blood vessel edge as the diameter of the blood vessel. When fitting the straight line using the least squares method, N pixels are used to fit the straight line each time, and the value of N is 5.

[0065] (2.4) During coronary angiography, contrast agent is introduced into the blood vessels through a catheter, making the vessels visible in the image. Therefore, the catheter is the starting point of the blood vessels in each image, located in the upper left corner. A tri-tree vascular tree based on spatial location is established using the angiography catheter as the starting point.

[0066] (2.5) Matching of blood vessels in coronary angiography sequences based on information of blood vessel segments.

[0067] The formula for extracting the intersection points of the vascular skeleton in (2.2) is as follows:

[0068]

[0069] In the formula, N(p1) represents the sum of the 8 neighboring pixels of pixel p1, and pi, (i = [2,3,...8,9]) represents the pixels arranged clockwise starting above p1.

[0070] The spatial location-based tritree vascular tree mentioned in (2.4) includes:

[0071] (2.4.1) Starting from the catheter, determine the child nodes of each vascular node sequentially, based on the Manhattan distance between the end coordinates of the node's vascular endpoint and the starting coordinates of the child node. The formula is:

[0072] d n =|x s -x e |+|y s -y e |

[0073] Here d n x represents the distance to Manhattan. s and y s The x-coordinate represents the starting coordinate of the child node. e and y e The endpoint coordinates of a node are represented by d. Child nodes, with the smaller distance from the endpoint coordinates, locate the starting coordinates of the blood vessels in the child nodes. n The threshold is 10;

[0074] (2.4.2) The position of a child node in the vascular tree is determined based on the spatial positions of multiple child nodes. When node p1(x1,y1) has two child nodes p2(x2,y2) and p3(x3,y3), the formula is as follows:

[0075] V=(x2-x1)×(y3-y2)-(y2-y1)×(x3-x2)

[0076] Here, when V is greater than 0, p1, p2, and p3 are arranged in counterclockwise order, so p2 is the first child node; otherwise, p3 is the first child node. When there are 3 child nodes, the arrangement order of the 3 child nodes is calculated by pairwise with the coordinates of the node's endpoint, and then the arrangement order of the 4 points is determined.

[0077] (2.4.3) Based on the growth characteristics of coronary angiography, assign an ID value to each vessel from top to bottom. Figure 4 The image shows the visualization of the constructed coronary artery tree.

[0078] The matching of vessels in the coronary angiography sequence in (2.5) includes:

[0079] (2.5.1) Calculate the coordinates S1(x) of the vessel origin in the preceding and following angiographic images. s1 ,y s1 ) and S2(x s2 ,y s2 The endpoint coordinates are E1(x). e1 ,y e1 ) and E2(x e2 ,y e2 The Manhattan distance of ) is calculated using the formula:

[0080] d s =|x s1 -x s2 |+|y s1 -y s2

[0081] d e =|x e1 -x e2 |+|y e1 -y e2 |

[0082] (2.5.2) The vessel length is obtained by calculating the sum of the pixels containing the vessel. The ratio of the vessel lengths L1 and L2 in the preceding and following angiographic images is calculated using the following formula:

[0083]

[0084] (2.5.3) Match blood vessels according to the conditional formula, which is:

[0085]

[0086] In the formula, the value of T is 65, and the value of R is 0.85. Figure 5 The image shown is a diagram of the coronary artery matching results.

[0087] The coronary artery myocardial bridging detection module of this invention includes:

[0088] (3.1) Calculate the degree of vascular stenosis using the following formula:

[0089]

[0090] In the formula, P represents the degree of vascular stenosis, and D... min D represents the minimum diameter of a blood vessel. f and D b Represents D min Normal diameter values ​​for both sides of the blood vessels; when the blood vessels are peripheral vessels, D min This represents the local minimum value of a blood vessel. Stenosis is classified into three grades: 30% to 50% is mild stenosis, 50% to 70% is moderate stenosis, and more than 70% is severe stenosis. Figure 6 This is a schematic diagram of stenosis detection;

[0091] (3.2) Based on the changes in the degree of vascular stenosis between consecutive frames, myocardial bridging is detected using the following formula:

[0092]

[0093] In the formula, stenosis_max and stenosis_min represent the maximum and minimum values ​​of vascular stenosis in the angiographic sequence, respectively; stenosis_max_frame and stenosis_min_frame represent the positions of the extreme values ​​of vascular stenosis in the sequence, respectively; and T... s and T f T represents the threshold of stenosis and the distance to the extreme value of vascular stenosis, respectively. s The threshold is 0.25, T f The value is 3;

[0094] (3.3) In actual cases, there may be cases of overall vascular stenosis, making it impossible to calculate the degree of stenosis. In such cases, myocardial bridging can be detected by changes in the average diameter, using the following formula:

[0095]

[0096] In the formula, average_max and average_min represent the maximum and minimum mean diameters of blood vessels in the angiography sequence, respectively; average_max_frame and average_min_frame represent the positions of the maximum and minimum values ​​in the sequence, respectively; and T... a The threshold T represents the ratio of the minimum to the maximum value. a The value is 0.75. Figure 7 This is the result of myocardial bridging detection.

Claims

1. A device for detecting myocardial bridging in a coronary angiography sequence, comprising: A coronary artery segmentation module is used to segment coronary arteries using a neural network model. This model consists of CNN and transformer structures, and can effectively extract local and global information about the blood vessels. The neural network model includes: (1.1) The neural network model is an end-to-end model. The encoding module consists of a residual module, an improved transformer encoding module, and a max pooling layer. The decoding module consists of a residual module, an improved transformer decoding module, a deconvolution layer, and skip connections. The improved transformer codec structure formula is as follows: ； In the formula, and It is obtained by expanding a small-sized feature map, which is smaller than the original feature map, through two different 1x1 convolutions and downsampling operations; (1.2) The coding module consists of a 5-layer structure, with each layer connected by a max pooling layer. The first layer consists of 2 residual modules, and the other layers consist of 1 residual module and 1 improved transformer coding module. (1.3) The decoding module is also composed of a 5-layer structure. Except for the 5th layer, each of the other layers has a residual module. The 2nd, 3rd and 4th layers achieve feature map upsampling and feature fusion of the corresponding layer in the encoding module through the improved transformer decoding structure. The 1st and 2nd layers are connected by a deconvolution layer. At the same time, the 1st layer achieves feature fusion with the 1st layer in the encoding module through a skip connection. The coronary artery information calculation module is used to calculate information including the diameter, length, and origin-end coordinates of blood vessels based on the vessel segmentation map, build a coronary artery tree containing all blood vessels, and calculate matching vessels between angiography sequences; including: (2.1) Extract the vascular skeleton from the vascular segmentation map; (2.2) Extract the endpoints and intersections in the vascular skeleton, separate the blood vessels to obtain vascular segments, and remove the spiky blood vessels; (2.3) Record the pixel coordinates at both ends of the blood vessel segment as the starting and ending coordinates of the blood vessel, calculate the number of pixels in the blood vessel segment as the blood vessel length, fit the blood vessel straight line using the least squares method, then obtain the blood vessel normal, and then calculate the distance between the intersection of the blood vessel normal and the blood vessel edge as the diameter of the blood vessel. (2.4) Starting with the angiography catheter, establish a trigeminal vascular tree based on spatial location; (2.5) Matching of blood vessels in coronary angiography sequences based on information about the vessel segments; The coronary myocardial bridging detection module is used to detect coronary myocardial bridging by calculating the degree of stenosis of the same vessel in the angiography sequence.

2. The myocardial bridging detection device in a coronary angiography sequence according to claim 1, characterized in that, In (2.1), the Zhang-Suen method is used to extract the vascular skeleton in the vascular segmentation map.

3. The myocardial bridging detection device in a coronary angiography sequence according to claim 1, characterized in that, The method for extracting the vascular skeleton intersection point in (2.2) is as follows: for a certain pixel point, if the sum of its 8 neighboring pixels is ≥3 and the sum of the pixels of its three adjacent neighboring pixels is ≤1, then this pixel point is judged as a vascular skeleton intersection point.

4. The myocardial bridging detection device in a coronary angiography sequence according to claim 1, characterized in that, The tritree vascular tree based on spatial location, as described in (2.4), includes: (2.4.1) Starting from the angiography catheter, determine the child nodes and their number for each vascular node in sequence, and use the Manhattan distance between the coordinates of the node's endpoint and the coordinates of the child node's starting point. To determine the child nodes of a node, the coordinates of the child node whose origin is less distant from the coordinates of the node's endpoints are taken as the child node's origin coordinates. Less than or equal to the threshold When the node is a child node, it is a child node of the node; otherwise, it is not a child node. (2.4.2) Determine the position of a child node in the vascular tree based on the spatial positions of multiple child nodes. When a node has two child nodes, determine the endpoint coordinates of the node. and the starting coordinates of the two child nodes and Form two vectors and The cross product of two vectors is given by the formula: ； Here For the cross product of vectors, when Greater than 0, , and Arranged in counter-clockwise order, therefore It is the first child node, otherwise It is the first child node; When there are 3 child nodes, the coordinates of the node's endpoint are... and the starting coordinates of the 3 child nodes , , Forming three pairs of vectors and , and , and Calculate the cross product of the three pairs of vectors respectively, and determine the order of the four points based on the value of the cross product; (2.4.3) Based on the growth characteristics of coronary angiography, assign an ID value to each vessel from top to bottom.

5. The myocardial bridging detection device in a coronary angiography sequence according to claim 1, characterized in that, The matching of vessels in the coronary angiography sequence in (2.5) includes: (2.5.1) For blood vessels in the previous frame image and blood vessels in the current frame image The coordinates of the origin and end points of the two blood vessels are respectively and , and Calculate the coordinates of the origin of blood vessels in the preceding and following angiographic images. and End point coordinates and Manhattan distance and ; (2.5.2) Assume blood vessels and The lengths are respectively and That is, calculate the ratio of blood vessel lengths in consecutive frames of angiography: ； (2.5.3) Determine the distance threshold and proportional threshold Match blood vessels according to the conditional formula, which is: 。 6. The myocardial bridging detection device in a coronary angiography sequence according to claim 1, characterized in that, The coronary myocardial bridging detection module includes: (3.1) Calculate the degree of vascular stenosis using the following formula: ； In the formula Indicates the degree of vascular stenosis. Represents the minimum diameter of the blood vessel. and represent The normal diameter of the blood vessels on both sides; when the blood vessels are peripheral vessels, Represents the local minimum value of a blood vessel; (3.2) Based on the changes in the degree of vascular stenosis between consecutive frames, when the change in the degree of stenosis exceeds a threshold At that time, it was diagnosed as myocardial bridging; (3.3) In cases of overall vascular stenosis, myocardial bridging is detected by changes in the average diameter.

Images