CN117934651B - Coronary artery CTA projection transformation method for lesion coronary artery detection - Google Patents

Abstract

The application discloses a coronary artery CTA projection transformation method for lesion coronary artery detection, which comprises the following steps: acquiring a CCTA pretreatment image based on a CTA image sequence of a patient coronary artery; performing automatic AI segmentation to obtain a three-dimensional model; constructing a rotation matrix and enabling the three-dimensional model to rotate by a set angle to obtain two-dimensional blood vessel images under projections of different body positions; preprocessing the two-dimensional blood vessel image to obtain a second preprocessed image; based on the second preprocessed image, a geometric quantitative feature index parameter is obtained. The fractal dimension and the curvature measured under different projections are quantitative indexes without units, and cannot be influenced by image factors and individual factors of patients. The application combines the multiple-transformation multi-view observation method, can provide more clinical and scientific research values than single projection, and can find the defect of insufficient plaque exposure on a single body position. Improving the accuracy and sensitivity of diagnosis and providing important help for doctors in treatment.

Description

Coronary artery CTA projection transformation method for lesion coronary artery detection

Technical Field

The application belongs to the technical field of image analysis, and particularly relates to a coronary artery CTA projection transformation method for lesion coronary artery detection.

Background

Coronary heart disease is a common cardiovascular disease and severely threatens human health. It is mainly caused by atherosclerotic plaques of coronary vessels, which may lead to vascular stenosis, thus causing serious consequences such as myocardial ischemia, myocardial infarction, etc. With the continuous development of medical technology, vascular detection technology has become an integral part of clinical medicine. Among them, detection of coronary vascular lesions is one of the important means for preventing and treating coronary heart disease. Currently, the common coronary vessel detection methods in clinic include coronary angiography, computed Tomography Angiography (CTA), and the like. However, coronary angiography has some disadvantages as a "gold standard" for diagnosing coronary heart disease, for example DSA is an invasive examination, and complications may exist. Therefore, to overcome the invasive examination, the present invention starts to explore a vascular detection method based on CCTA projective transformation. According to the method, the CCTA image data is utilized to conduct AI automatic blood vessel segmentation extraction and projection transformation, so that non-invasive examination can be achieved, complications are avoided, clearer and more visual blood vessel images can be obtained, and the accuracy and reliability of blood vessel detection are improved.

In recent years, a coronary plaque automatic detection method based on deep learning has received a great deal of attention. IVANCEVIC et al propose an automatic coronary plaque detection method based on a deep convolutional neural network. The method uses a deep neural network to classify and locate plaques by segmenting and extracting features of coronary CTA images. The method obtains higher accuracy and recall rate on a plurality of data sets, and provides an effective solution for automatic detection of coronary plaque. However, existing methods tend to focus on patch detection at a single image level only, ignoring the continuity and uniformity of patches across multiple image levels. This may lead to a condition of plaque omission or false detection.

Disclosure of Invention

The application provides a coronary artery CTA projection transformation method for lesion coronary artery detection, which aims to solve the technical problems that the prior art only focuses on plaque detection on a single image level and ignores continuity and uniformity of plaque on a plurality of image levels.

In order to solve the technical problems, the application adopts a technical scheme that: a coronary CTA projective transformation method for lesion coronary detection, comprising the steps of:

acquiring a CCTA pretreatment image based on a CTA image sequence of a patient coronary artery;

AI automatic segmentation is carried out based on the CCTA pretreatment image, and a three-dimensional model is obtained;

Constructing a rotation matrix and enabling the three-dimensional model to rotate by a set angle to obtain two-dimensional blood vessel images under projections of different body positions;

Preprocessing the two-dimensional blood vessel image to obtain a second preprocessed image;

Based on the second preprocessed image, a geometric quantitative feature index parameter is obtained.

Further, the method for automatically dividing AI based on CCTA preprocessing image to obtain the three-dimensional model comprises the following steps:

based on the CCTA preprocessing image and the 3D UX-Net model, acquiring a three-dimensional model of the segmented coronary artery;

Based on the three-dimensional model of the segmented coronary artery, the blood vessels of the non-region of interest are hidden/exposed.

Further, the method for constructing a rotation matrix and rotating the three-dimensional model by a set angle to obtain two-dimensional blood vessel images under different body position projections comprises the following steps:

Based on DSA common radiography position and basic position AP position, rotating the three-dimensional model of the segmented coronary artery by taking LR, PA and I S as axes; the body position rotation angle is specifically as follows: the right shoulder IS at least one of LR-30 degrees, the left shoulder IS I S-30 degrees +LR-20 degrees, the spider IS at least one of IS-45 degrees +LR30 degrees, the foot IS at least one of LR30 degrees, the liver IS at least one of IS-30 degrees +LR20 degrees, the right shoulder IS at least one of IS-30 degrees +LR-20 degrees, the left front oblique IS at least one of IS-45 degrees, and the head IS at least one of LR-20 degrees.

Further, the method for constructing the rotation matrix comprises the following steps:

Based on the three-dimensional model of the segmented coronary artery, adjusting the AP position of the basic body position; the rotation angle of the three-dimensional model around an x-axis (LR) is recorded as alpha, and a first rotation matrix Rx is obtained; the rotation angle of the three-dimensional model around the y axis (PA) is recorded as beta, and a second rotation matrix Ry is obtained; the rotation angle of the three-dimensional model around the z axis (IS) IS marked as gamma, and a third rotation matrix Rz IS obtained;

Acquiring a final rotation matrix R based on the first rotation matrix Rx, the second rotation matrix Ry and the third rotation matrix Rz;

Based on the three-dimensional model rotated according to different angles, two-dimensional blood vessel images under different projections are obtained.

Further, fractal dimension measurement is performed on a vascular network in the two-dimensional vascular image, and curvature measurement is performed on a coronary artery trunk in the two-dimensional vascular image.

Further, the method for obtaining the final rotation matrix R based on the first rotation matrix Rx, the second rotation matrix Ry and the third rotation matrix Rz includes:

Acquiring the first rotation matrix Rx based on the formula (1); wherein, the formula (1) is:

Acquiring the second rotation matrix Ry based on the formula (2); wherein, the formula (2) is:

acquiring the third rotation matrix Rz based on the formula (3); wherein, the formula (3) is:

Acquiring a final rotation matrix R based on the formula (4); wherein, the formula (4) is:

R＝Rx*Ry*Rz (4)。

The coronary artery CTA projection transformation method is applied to the preparation of products for detecting vascular coronary artery plaque.

The beneficial effects of the application are as follows: the fractal dimension and the curvature in the application are no unit indexes, and are not influenced by image factors and individual factors of patients. The method can automatically extract the coronary three-dimensional model, and improves the working efficiency and the model precision. The method can observe coronary plaque at multiple angles through multiple transformations, and the multiple transformations can provide more value than single projection, and can find the defect of insufficient plaque exposure on a certain body position. The method improves the accuracy and reliability of blood vessel detection, and is suitable for preventing and diagnosing coronary heart disease.

Drawings

FIG. 1 is a flow chart of an embodiment of a coronary CTA projective transformation method for lesion coronary detection of the present application;

FIG. 2 is a flowchart illustrating an embodiment of the step S2 in FIG. 1;

FIG. 3 is a schematic illustration of an automatic segmentation process of coronary vessels in a coronary CTA projective transformation method for lesion coronary detection of the present application;

FIG. 4 is a flowchart illustrating an embodiment of the step S3 in FIG. 1;

FIG. 5 is a projection of the segmented coronary based three-dimensional model of the present application after rotation in the forward position;

FIG. 6 is a projection of a segmented coronary based three-dimensional model of the present application after rotation of the spider bits;

FIG. 7 is a projection of the segmented coronary based three-dimensional model of the present application after rotation at the liver site;

FIG. 8 is a projection view of the second pre-treatment of FIG. 5;

FIG. 9 is a projection view of the second pre-treatment of FIG. 6;

FIG. 10 is a projection view of the second pre-treatment of FIG. 7;

FIG. 11 is a schematic illustration of a coronary artery in a body position; wherein, FIG. 11a shows no lesions of the coronary artery; FIG. 11b is a schematic illustration of a coronary artery in another body site exposing a portion of plaque; figure 11c shows that the figure position can fully expose the plaque.

Detailed Description

The present invention will be described in further detail with reference to specific examples in order to make the objects, technical solutions and advantages of the present invention more apparent.

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced otherwise than as described herein, and therefore the present invention is not limited to the specific embodiments of the disclosure that follow.

Referring to fig. 1, fig. 1 is a flowchart illustrating an embodiment of a coronary CTA projective transformation method for lesion coronary detection according to the present application. The method comprises the following steps:

s1, acquiring a CCTA preprocessing image based on a CTA image sequence of a patient coronary artery.

Specifically, a patient acquires a coronary artery CTA image sequence of the patient through Computed Tomography Angiography (CTA), and performs preprocessing to obtain a CCTA preprocessed image.

And S2, carrying out AI automatic segmentation based on the CCTA preprocessing image to obtain a three-dimensional model.

Specifically, referring to fig. 2, fig. 2 is a flow chart illustrating an embodiment of step S2 in fig. 1. The specific method of the step S2 comprises the following steps:

S21, acquiring a three-dimensional model of the segmented coronary artery based on the CCTA preprocessing image and the 3D UX-Net model.

Specifically, the 3D UX-Net model processing involves a pre-training stage using labeled vessel and plaque datasets, randomly cropping and enhancing it, and encoding it using Swin UNETR. As shown in fig. 3, mask volume rendering, contrast learning, and rotation prediction are employed as proxy tasks for learning the input image context representation during the pre-training process.

The method comprises the following specific steps: a blood vessel feature representation layer is created to convert the blood vessel image data into a form capable of representing various features of the blood vessel. For each sub-stripe, weights are determined by calculating their average gray value and connectivity, and weight initialization is performed. Each vessel feature is then SKIPLIST coded to distinguish them from the surrounding background. The encoded sub-bands are arranged in a left to right, top to bottom order to better distinguish between blood vessels and surrounding tissue. After the blood vessel feature representation layer is established, a region detection layer needs to be established. The function of this layer is to pre-process the vessel feature representation layer to extract the region detection features of the vessel. Next, the blood vessel feature representation layer and the region detection layer are imported into the fully connected layer to enable connectivity detection of the blood vessel. The fully connected layer will use the entered features for calculations and analysis to determine the connectivity of the vessel. Then, the output of the region detection layer and the output of the full connection layer are input into a decision layer to realize classification and labeling of blood vessels. The decision layer judges and decides the input characteristics to determine the category of the blood vessel and carry out corresponding labeling work. Next, a vessel segmentation model needs to be trained. The vessel segmentation model is input into a decision layer to segment and label the vessel. The method comprises the steps of accurately segmenting and labeling blood vessels through learning and training of a model. In order to verify the accuracy and the robustness of the blood vessel segmentation model, manual verification is performed.

Step S22, hiding/exposing the non-interested area and the blood vessel of the interested area based on the three-dimensional model of the segmented coronary artery.

Specifically, based on the three-dimensional model of the segmented coronary artery that has been formed, the blood vessels that hide the non-region of interest and the region of interest may be selected, and the blood vessels of the non-region of interest and the region of interest may also be exposed.

S3, constructing a rotation matrix and enabling the three-dimensional model to rotate by a set angle to obtain two-dimensional blood vessel images under projection of different body positions.

Specifically, referring to fig. 4, fig. 4 is a flow chart illustrating an embodiment of step S3 in fig. 1. The specific method of the step S3 comprises the following steps:

s31, rotating the three-dimensional model of the segmented coronary artery by taking LR, PA and IS as axes based on the DSA common radiography position and the basic position AP.

Specifically, referring to fig. 5, the rotation angle of the three-dimensional model about the x-axis (LR) is denoted as α, where the rotation matrix about the x-axis (LR) rotated by α is as shown in formula (1):

the rotation angle of the three-dimensional model around the y-axis (PA) is recorded as beta; the rotation matrix rotated by β angle around the y-axis (PA) is as shown in equation (2):

the rotation angle of the three-dimensional model around the z axis (IS) IS denoted as gamma, and the rotation matrix around the z axis (IS) rotated by the gamma angle IS as shown in formula (3):

Step s32, the final rotation matrix R of the coronary model may be obtained by multiplying the three rotation matrices, where the final rotation matrix R is shown in formula (4):

R＝Rx*Ry*Rz (4)；

The rotation angles of all the body positions are specifically as follows: right head (LR-30 °), left shoulder (IS-30 ° + LR-20 °), spider (IS-45 ° + LR30 °), foot (LR 30 °), liver (IS 30 ° + LR-20 °), right shoulder (IS 30 ° + LR-20 °), left front oblique (IS-45 °), head (LR-20 °).

And S33, obtaining two-dimensional blood vessel images under different projections based on the models rotated at different angles.

Specifically, as shown in fig. 6-8, the non-interested region may be hidden when the two-dimensional image is taken, for example, the right crown may be hidden when the left crown two-dimensional image is taken.

S4, preprocessing the two-dimensional blood vessel image to obtain a second preprocessed image.

Specifically, the obtained two-dimensional image is subjected to preprocessing such as binarization and skeletonization.

Referring to fig. 9 to 11, the two-dimensional image obtained in step S3 is first converted into 8-bit and the initially set background is set to black to prevent the image from becoming INVERTING LUT.

Secondly, each pixel point of the image is given a gray value of 0 (black) or 255 (white), so that the whole image presents obvious black-and-white effect as only two colors of black and white. An appropriate threshold is selected to distinguish the background from the foreground of the image, and for each pixel it is set to white (255) if its grey value is greater than or equal to the threshold, or else to black (0). After a proper threshold value is selected, all the blood vessel segments are made to be white images, and the background is made to be black images, so that the image binarization is realized.

The intensity value of each pixel is then replaced with the average value of the pixels in its neighboring 3 x 3 region to reduce noise in the image and smooth the surface of the image. And then re-binarizing the resultant.

Finally, skeletonizing, which is a morphological skeletonizing algorithm. Including corrosion, swelling, and repeated iterations. Background pixels surrounding each foreground pixel are found first and these pixels are deleted, making the foreground pixels "thin". And expanding the image after the etching operation. And repeatedly carrying out corrosion and expansion operation until the skeletonization of the image reaches a stable state, and obtaining the single-pixel broadband skeleton.

The steps can be recorded by using macro codes to construct a macro file, so that repeated operations are reduced, and the image processing efficiency is improved.

Holes, disconnected edges, etc. can be filled appropriately based on the skeletonized image.

S5, acquiring geometric quantitative characteristic index parameters based on the second preprocessed image.

Specifically, based on the second preprocessing image, geometric quantitative characteristic index parameters such as fractal dimension, coronary artery trunk curvature and the like are measured.

The method for measuring the geometric quantitative characteristic index parameter comprises the following steps: firstly, marking all pixels in a skeleton image based on the skeletonized image obtained in the step S4, and calculating parameters such as all connection points and branches of the skeleton, the length of the skeleton, the number of branches and the like.

Specifically, for regular fractal, the fractal dimension of the vascular skeleton can be calculated directly by using a Scaling rule: d represents the fractal dimension, N represents the number of units of length, ε represents the reduction of the units of length compared to the original.

For irregular shapes, a fractal Box dimension (Box-counting dimension) was introduced, which is based on the concept of Box counting (Box-counting), the complexity of which is measured by calculating the number of boxes covering the fractal object.

The image obtained in step S4 is filled into a series of boxes. The number of points in each block for which the pixel value is not 0 is counted. The complexity of the image in this area may be reflected.

The number of pixel points in each box is counted and divided by the side length of the box to obtain the box dimension of the box filling image. The fractal characteristics of the image in that region may be reflected.

The fractal dimension of the image can be obtained by fitting the box dimension.

Based on fractal dimension measurement, the complexity of the vascular network can be reflected, and whether the neovascularization exists or not can be observed.

And (3) measuring the curvature of the coronary artery trunk based on the preprocessed image obtained in the step S4.

The scale is then set and each pixel on the curve image is manually selected, and the second derivative value of the input surface, i.e. the surface curvature at that pixel, is calculated pixel by pixel.

Next, an ideal fitted curve or model is assumed using the selected location of each pixel.

A loss function is defined, and the difference between the fitted curve and the actual observed data is measured.

An optimization algorithm is used to find the parameter values that minimize the loss function.

Finally, the best fit Curve is obtained, and the Avg. Curve Length, curvature sigma ² and Point Curvature indexes of the Curve can be obtained.

Referring to fig. 11, fig. 11a shows no coronary lesions; FIG. 11b is a schematic illustration of a coronary artery in another body site exposing a portion of plaque; figure 11c shows that the figure position can fully expose the plaque. By way of example, a coronary heart disease patient is selected, coronary CTA examination is performed, and the following diagnosis results are obtained by detailed analysis of the images:

Left Anterior Descending (LAD) proximal vessel wall mixed plaque, lumen severe stenosis; the middle tube wall calcified plaque and the lumen was moderately narrow. The first diagonal branch (D1) has calcified plaque on the wall of the vessel, and the vessel cavity is slightly narrow. Left Circumflex (LCX) proximal vessel wall multiple mixed plaque, lumen severe stenosis. Calcified plaque of the wall of the proximal segment and middle segment of the Right Coronary Artery (RCA) and slight stenosis of the lumen; the distal tube wall calcifies plaque and the lumen is slightly stenosed. The geometric characteristics of the lesion coronary with the atherosclerosis plaque are analyzed in this example, 8 common contrast body positions are simulated according to the quantitative index measurement method, and the measurement results are shown in table 1.

TABLE 1 quantitative coronary index measurement results for patients with coronary heart disease

The method provides a method for observing coronary artery with multiple indexes and multiple dimensions, and has important significance for preventing and diagnosing coronary heart disease.

The method simulates the common body position of coronary angiography, adopts a method of multiple transformation and multiple visual angles, can find out more multivalent values than single projection, and finds out the defect of insufficient plaque exposure on a certain single body position; on the basis, the fractal dimension measurement is carried out on the vascular network, and quantitative indexes such as the main curvature of the coronary artery are measured, so that the defect that the coronary artery has only qualitative indexes is overcome. The method of the application ensures that the observation result is more accurate and can better serve clinicians.

The foregoing description is only illustrative of the present application and is not intended to limit the scope of the application, and all equivalent structures or equivalent processes or direct or indirect application in other related technical fields are included in the scope of the present application.

Claims (2)

1. The coronary artery CTA projection transformation method for detecting pathological coronary artery is characterized by comprising the following steps of:

acquiring a CCTA pretreatment image based on a CTA image sequence of a patient coronary artery;

Performing AI automatic segmentation based on the CCTA pretreatment image to obtain a three-dimensional model; the method for acquiring the three-dimensional model comprises the following steps: acquiring a three-dimensional model of the segmented coronary artery based on the CCTA pretreatment image and the 3DUX-Net model; hiding/exposing a non-region of interest and a vessel of the region of interest based on the three-dimensional model of the segmented coronary artery;

Constructing a rotation matrix and enabling the three-dimensional model to rotate by a set angle to obtain two-dimensional blood vessel images under projections of different body positions; based on a DSA common contrast body position and a basic body position AP position, rotating the three-dimensional model of the segmented coronary artery by taking LR, PA and IS as axes; the body position rotation angle is specifically as follows: the right shoulder position IS LR-30 degrees, the left shoulder position IS IS-30 degrees+LR-20 degrees, the spider position IS IS-45 degrees+LR30 degrees, the foot position IS LR30 degrees, the liver position IS IS30 degrees+LR20 degrees, the right shoulder position IS IS30 degrees+LR20 degrees, the left front oblique position IS IS-45 degrees, and the head position IS LR-20 degrees;

The method for constructing the rotation matrix comprises the following steps: based on the three-dimensional model of the segmented coronary artery, adjusting an AP position of the basic body position; the rotation angle of the three-dimensional model around x is marked as alpha, and a first rotation matrix Rx is obtained; the rotation angle of the three-dimensional model around the y-axis is recorded as beta, and a second rotation matrix Ry is obtained; the rotation angle of the three-dimensional model around the z axis is marked as gamma, and a third rotation matrix Rz is obtained; acquiring a final rotation matrix R based on the first rotation matrix Rx, the second rotation matrix Ry and the third rotation matrix Rz; based on the three-dimensional model rotated according to different angles, obtaining two-dimensional blood vessel images under different projections;

The method for obtaining the final rotation matrix R based on the first rotation matrix Rx, the second rotation matrix Ry and the third rotation matrix Rz includes:

acquiring the first rotation matrix Rx based on formula (1); wherein, the formula (1) is:

Acquiring the second rotation matrix Ry based on formula (2); wherein, the formula (2) is:

acquiring the third rotation matrix Rz based on the formula (3); wherein, the formula (3) is:

acquiring the final rotation matrix R based on the formula (4); wherein, the formula (4) is:

R＝Rx*Ry*Rz(4)；

preprocessing the two-dimensional blood vessel image to obtain a second preprocessed image;

and acquiring geometric quantitative characteristic index parameters based on the second preprocessed image.

2. The method of claim 1, wherein fractal dimension measurements are performed on a network of vessels in the two-dimensional vessel image and curvature measurements are performed on a coronary trunk in the two-dimensional vessel image.