Early pulmonary artery micro embolism detection method based on angiography CT (Computed Tomography) image

Abstract

The invention provides an angiography CT (Computed Tomography) image-based early-stage pulmonary artery micro embolism detection method, which comprises the following steps of: extracting pulmonary parenchyma, and segmenting a small pulmonary blood vessel to obtain a pulmonary blood vessel tree; network weight is updated based on a neural evolution calculation method, and lung arteries and veins are separated; and obtaining a complete pulmonary artery tissue from a pulmonary artery and vein separation result, establishing a level set evolution model to enable the curve to be accurately evolved to the edge of a normal pulmonary artery tissue, and removing the normal pulmonary artery tissue to obtain a pulmonary embolism focus tissue so as to obtain a pulmonary embolism detection result. The method provided by the invention can accurately detect the micro embolism part at the early stage of pulmonary embolism, and has important clinical application value.

Description

Technical Field

[0001] The invention belongs to the technical field of medical image segmentation methods, and in particular relates to an early pulmonary artery microembolism detection method based on angiographic CT images. Background Art

[0002] Pulmonary embolism (PE), also known as pulmonary embolism, is a clinical syndrome in which endogenous or exogenous emboli block the main trunk of the pulmonary artery or its branches, causing pulmonary circulation disorders. After thrombus embolism in the pulmonary artery, the thrombus is insoluble and organized, leading to increased pulmonary vascular resistance and progressive increase in pulmonary artery pressure, eventually causing right ventricular hypertrophy and right heart failure. In recent years, this disease has become an ischemic disease second only to hypertension and heart disease, and its incidence rate has shown an increasing trend year by year in my country. Clinically, the mortality rate of patients with misdiagnosis and missed diagnosis is as high as 30%, but the mortality rate of patients who are discovered and treated in time can be reduced by 10 times. In the early stages of pulmonary embolism, effective diagnosis and treatment can effectively reduce the mortality rate of pulmonary embolism.

[0003] At present, there are many imaging examination methods for clinical diagnosis of PE, including CT, radionuclide imaging, ultrasound, pulmonary angiography, etc. Different examination methods have their own advantages and disadvantages. X-ray chest films cannot directly show thrombus, but can show some indirect signs of PE, including regional reduction of lung texture, increased lung translucency, elevated diaphragm and pleural effusion. Echocardiography can directly show thrombus and evaluate hemodynamic changes, but this detection is rare in transthoracic echocardiography, and only the main pulmonary artery and the left and right pulmonary artery trunks can be observed. Pulmonary angiography can not only show the location, range and degree of PE, but also obtain hemodynamic data. However, pulmonary angiography is expensive and relies on highly skilled professionals.

[0004] Computed Tomography Angiography (CTA) is an image reconstruction technique that performs CT scans of pulmonary angiography after intravenous injection of contrast agents. It can display vascular images in three dimensions and has become an effective way to diagnose PE in clinical practice. However, CTA images usually contain hundreds of images, and the images contain various PE-like objects (false positives), such as image artifacts, partial volume artifacts, lymph nodes, and vascular bifurcations. Therefore, manual reading of images is a very laborious, time-consuming, and complicated task. Computer-aided detection (CAD) can help radiologists save a lot of reading time, so it is very meaningful to design a fully automatic pulmonary embolism detection method that effectively improves the PE detection rate while reducing the missed diagnosis rate of PE. Summary of the invention

[0005] In view of this, the object of the present invention is to provide an early pulmonary artery microembolism detection method based on angiographic CT images, so as to accurately identify microemboli in the early stage of pulmonary embolism.

[0006] The technical solution of the present invention is: a method for detecting early pulmonary artery microembolism based on angiographic CT images, comprising:

[0007] Extract lung parenchyma, segment small pulmonary vessels, and obtain the pulmonary vascular tree;

[0008] Based on the pulmonary vascular tree, the network weights are updated based on the neuroevolutionary computing method to separate the pulmonary arteries and veins, and accurate pulmonary artery and vein separation results are obtained;

[0009] Complete pulmonary artery tissue is obtained from the pulmonary artery and vein separation results, and a level set evolution model is established to allow the curve to accurately evolve to the edge of normal pulmonary artery tissue. The pulmonary embolism lesion tissue is obtained after removing the normal pulmonary artery tissue, that is, the detection result of pulmonary embolism is obtained.

[0010] Furthermore, the extraction of lung parenchyma and segmentation of small pulmonary vessels to obtain a pulmonary vascular tree includes:

[0011] The entire chest cavity tissue of the lung region is extracted using region growing to obtain the chest cavity region;

[0012] Use morphological closing to remove cavities in the thoracic region caused by blood vessels and other tissues;

[0013] The fractional differential is then used to increase the contrast of the blood vessels.

[0014] Furthermore, the method of updating the network weights based on the neuroevolutionary computing method to separate the pulmonary arteries and veins to obtain accurate pulmonary artery and vein separation results includes:

[0015] The three-dimensional pulmonary vascular data of the pulmonary vascular tree is expanded with a small sample data to obtain two-dimensional images of the pulmonary vessels in different directions;

[0016] Establish a convolutional neural network model and use the Cauchy mutation evolution algorithm based on clonal selection and gene position optimization to update the network parameters;

[0017] Finally, accurate pulmonary artery and vein separation results are obtained by training the network;

[0018] Furthermore, the three-dimensional pulmonary vascular data of the pulmonary vascular tree is expanded by small sample data to obtain two-dimensional images of the pulmonary vessels in different directions: the three-dimensional pulmonary vascular data is sampled and decomposed into sub-blocks of the same size, and the obtained sub-blocks are cut in different directions according to the center points to obtain two-dimensional images of the blood vessels in different directions.

[0019] Furthermore, the level set evolution model is established to make the curve evolve accurately to the edge of normal pulmonary artery tissue, and the pulmonary embolism lesion tissue is obtained after removing the normal pulmonary artery tissue, including:

[0020] Based on the pulmonary artery data in the pulmonary artery and vein separation results, the mean and variance information of the pulmonary artery is counted and a pulmonary artery grayscale information model is established;

[0021] According to the pulmonary artery grayscale information model, a length constraint item, a grayscale constraint item and an image information item are obtained;

[0022] The knot length constraint term and image information term are used to establish the level set evolution model, and the embolized tissue is obtained through the level set curve evolution.

[0023] The present invention provides a method for detecting early pulmonary artery microembolism based on angiographic CT images, which includes pulmonary vascular segmentation, pulmonary artery and vein separation, and pulmonary artery embolism detection, solves the high-performance computing problem of image processing, and has important theoretical and academic research value.

[0024] The method provided by the present invention can be integrated into the medical auxiliary diagnosis system to assist doctors in diagnosing related diseases, and has clear social value and application value. With the increasing understanding of pulmonary embolism in clinical practice, how to accurately identify the location of embolism in the early stage of pulmonary embolism and take active and effective intervention has important clinical value and scientific significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 The example of pulmonary embolism in CTA images provided by the present invention is shown. Figure 1 The white arrow in the middle indicates the pulmonary embolism lesion area;

[0028] Figure 2 A schematic diagram of the framework of the method for detecting early pulmonary artery microembolism based on angiographic CT images provided by the present invention;

[0029] Figure 3Schematic diagram of lung parenchyma segmentation provided by the present invention; wherein (a) the original lung image; (b) the chest cavity and background area obtained after region growth and hole filling; (c) the lung parenchyma extraction result;

[0030] Figure 4 A flow chart of the pulmonary artery and vein separation algorithm provided by the present invention;

[0031] Figure 5 This is a flow chart of pulmonary embolism detection provided by the present invention. DETAILED DESCRIPTION

[0032] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present invention. Instead, they are merely examples of systems consistent with some aspects of the present invention as detailed in the appended claims.

[0033] CTA images usually contain hundreds of images, and the images contain various PE-like objects (false positives), such as image artifacts, partial volume artifacts, lymph nodes, vascular bifurcations, etc. Therefore, manual reading is a very laborious, time-consuming and complicated task. Figure 1 Several cases of pulmonary embolism based on CTA images are shown. As can be seen from the figure, Figure 1 (a) and Figure 1 (b) shows the location of pulmonary embolism in a two-dimensional perspective. Pulmonary embolism often occurs in the pulmonary artery and can occur at any location in the artery with different shapes and sizes. In complex actual clinical practice, due to interference factors such as noise, motion artifacts, and imaging quality, the grayscale values ​​of the embolism site in some images overlap with those of tissues such as cartilage and lung wall, increasing the difficulty of detection.

[0034] Early detection methods of pulmonary artery microembolism based on angiographic CT images include:

[0035] Step 1: Extract lung parenchyma, segment small pulmonary vessels, and obtain the pulmonary vascular tree;

[0036] In the lung CTA images, the thicker parts of the pulmonary vascular branches are effectively enhanced after angiography, and have a certain contrast with the surrounding tissues, but the enhancement effect of the small blood vessels is limited. The above steps can effectively detect all the small pulmonary blood vessels, laying the foundation for the subsequent separation of pulmonary arteries and veins and arterial embolism detection; the method provided by the present invention is not affected by factors such as motion artifacts and noise, overcomes the problem of confusion with the lung parenchyma, and realizes the accurate segmentation of small pulmonary blood vessels; the present invention uses lung parenchyma extraction to segment the blood vessels, and density-based segmentation methods such as grayscale thresholding and region growing can be used to extract the lung area and exclude lung nodules and high-density blood vessels. However, this will cause more pits in the lung area. In order to overcome the above shortcomings, the present invention uses the inverse operation of region growing on the lung area, and uses region growing to extract the entire chest tissue. Lung parenchyma area ( Figure 3 The chest cavity and background area in a) need to be removed, and the chest cavity area is extracted by region growing. Several seed points are randomly selected in the chest cavity area, and the chest cavity is extracted by region growing. Finally, the morphological closing operation is used to remove the holes caused by blood vessels and other tissues, and then the contrast of the blood vessel part is increased by fractional differential.

[0037] Step 2: Based on the pulmonary vascular tree, the network weights are updated using a neuroevolutionary computing method to separate the pulmonary arteries and veins, thereby obtaining accurate pulmonary artery and vein separation results;

[0038] In clinical practice, some data are affected by factors such as the dose of contrast agent and imaging quality, and the grayscale values ​​of the pulmonary veins and some pulmonary arteries overlap. Using this step to effectively separate the arteries and veins of the pulmonary vessels is of great guiding significance for subsequent arterial embolism detection; clinical images are affected by the dose of contrast agent, and there is an overlap between the grayscale values ​​of some pulmonary arteries and the grayscale values ​​of the pulmonary veins, which increases the difficulty of separating the arteries and veins. Therefore, in CTA images generated under non-ideal conditions (artifacts, volume effects, noise), the present invention can effectively separate the pulmonary arteries and pulmonary veins.

[0039] The arteries and veins in the pulmonary blood vessels are crisscrossed, and it takes a lot of time for radiologists to manually outline the gold standard data of arteries and veins, and there is less data that can be used for deep learning training. It is difficult to improve the accuracy of the network output by simply changing the traditional deep learning network structure or debugging parameters. Therefore, to solve complex segmentation problems such as pulmonary artery and vein separation, the present invention adopts a neuroevolutionary computing method to update the network weights. First, the three-dimensional pulmonary vascular data is sampled and decomposed into sub-blocks of the same size. The obtained sub-blocks are cut in different directions according to the center point, and two-dimensional images of the blood vessels in different directions can be obtained. A convolutional neural network model is established, and the network parameters are updated using the Cauchy mutation evolution algorithm based on clonal selection and gene position optimization. Finally, the accurate pulmonary artery and vein separation results are obtained by training the network. The algorithm flow chart is shown in Figure 4;

[0040] Step 3: Obtain complete pulmonary artery tissue from the pulmonary artery and vein separation results, establish a level set evolution model to allow the curve to accurately evolve to the edge of normal pulmonary artery tissue, and obtain pulmonary embolism lesion tissue after removing the normal pulmonary artery tissue, that is, obtain the detection result of pulmonary embolism.

[0041] The complex network structure of the pulmonary arteries and their varying sizes make pulmonary embolism detection more challenging than pulmonary nodule detection. Detection of tiny embolic points in the early stages of pulmonary embolism plays a vital role in the prevention and later treatment of the disease. This step can accurately detect tiny embolic parts in the early stages of pulmonary embolism.

[0042] In clinical images, the grayscale values ​​of the pulmonary embolism and the pulmonary veins overlap due to factors such as contrast agent dosage, imaging quality, and motion artifacts, which increases the difficulty of detecting early microemboli. Early detection of the site of embolism and timely treatment can effectively inhibit the progression of the disease and improve the patient's quality of life, which has great clinical application value. The present invention can accurately identify microemboli in the early stage of pulmonary embolism.

[0043] Pulmonary embolism can cause vascular stenosis in CTA images, and the contrast is usually lower than that of the surrounding pulmonary artery tissue. Radiologists need to check the pulmonary artery and its branches layer by layer for abnormalities. When the amount of data is large, missed diagnosis or even misdiagnosis is inevitable. Therefore, computer-aided diagnosis of pulmonary embolism is urgently needed to reduce the workload of doctors. Steps 1 and 2 have obtained complete pulmonary artery tissue. At this time, the pulmonary artery contains normal arterial tissue and embolic lesion tissue. To this end, this step adopts a pulmonary embolism lesion segmentation method based on grayscale constrained level set. By establishing a level set evolution model, the curve is accurately evolved to the edge of the normal pulmonary artery tissue. After removing this part of the tissue, the pulmonary embolism lesion tissue is obtained, thereby realizing automatic detection of pulmonary embolism.

[0044] Figure 5 This is the pulmonary embolism detection method for this step. First, the pulmonary artery tissue is obtained. Since pulmonary embolism usually only accounts for a small proportion of the pulmonary artery tissue, it is necessary to count the mean and variance information of the pulmonary artery and establish a pulmonary artery grayscale information model. According to this model, the grayscale value range of the pulmonary artery tissue can be roughly estimated and the grayscale constraint rule term is designed. Then, the level set evolution model is established by combining the length term and the image information term, and the embolic tissue is obtained through the evolution of the level set curve.

[0045] The level set evolution model adopts the DRLSE model, which consists of three parts: length constraint term, area constraint term and rule term.

[0046] The energy expression is:

[0047] E DRLSE (φ)=λ1R(φ)+λ2L(φ)+λ3A(φ) (1)

[0048]

[0049] A(φ)=∫ Ω HkDJ ε (-φ)dx (4)

[0050] Where λ1, λ2, λ3 are non-negative constants. Formulas (2), (3), and (4) are the rule term, length constraint term, and area constraint term, respectively. The rule term can maintain the continuity of the signed distance function, and this term can effectively avoid the level set reinitialization problem. The length term is used to constrain the length of the evolution curve. When the curve converges to the target edge, the value of the length term is minimized. At the same time, this term can also make the surface of the evolution curve smoother. The area term can accelerate the evolution of the curve. When the evolution curve is far from the target edge, this term can effectively increase the speed of the curve evolution.

[0051] H(x) and δ(·) in formulas (5) and (6) are the Heaviside function and its derivative, respectively, defined as:

[0052]

[0053]

[0054] According to formulas (5) and (6), formula (4) can be rewritten as:

[0055]

[0056] Using the gradient descent flow minimization formula (7), we can get:

[0057]

[0058] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These changes and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for detecting early pulmonary artery microembolism based on angiographic CT images, characterized in that: include: Extract lung parenchyma, segment small pulmonary vessels, and obtain the pulmonary vascular tree; Based on the pulmonary vascular tree, the network weights are updated based on the neuroevolutionary computing method to separate the pulmonary arteries and veins, and accurate pulmonary artery and vein separation results are obtained; Complete pulmonary artery tissue is obtained from the pulmonary artery and vein separation results, and a level set evolution model is established to allow the curve to accurately evolve to the edge of normal pulmonary artery tissue. The pulmonary embolism lesion tissue is obtained after removing the normal pulmonary artery tissue, that is, the detection result of pulmonary embolism is obtained.

2. The method for detecting early pulmonary artery microembolism based on angiographic CT images according to claim 1, characterized in that: The extraction of lung parenchyma and segmentation of small pulmonary vessels to obtain a pulmonary vascular tree includes: The entire chest cavity tissue of the lung region is extracted using region growing to obtain the chest cavity region; Use morphological closing to remove cavities in the thoracic region caused by blood vessels and other tissues; The fractional differential is then used to increase the contrast of the blood vessels.

3. The method for detecting early pulmonary artery microembolism based on angiographic CT images according to claim 1, characterized in that: The method of updating the network weights based on the neuroevolutionary computing method to separate the pulmonary arteries and veins and obtain accurate pulmonary artery and vein separation results includes: The three-dimensional pulmonary vascular data of the pulmonary vascular tree is expanded with a small sample data to obtain two-dimensional images of the pulmonary vessels in different directions; Establish a convolutional neural network model and use the Cauchy mutation evolution algorithm based on clonal selection and gene position optimization to update the network parameters; Finally, accurate pulmonary artery and vein separation results are obtained by training the network.

4. The method for detecting early pulmonary artery microembolism based on angiographic CT images according to claim 3, characterized in that: The three-dimensional pulmonary vascular data of the pulmonary vascular tree is expanded with small sample data to obtain two-dimensional images of the pulmonary vessels in different directions: the three-dimensional pulmonary vascular data is sampled and decomposed into sub-blocks of the same size, and the obtained sub-blocks are cut in different directions according to the center points to obtain two-dimensional images of the blood vessels in different directions.

5. The method for detecting early pulmonary artery microembolism based on angiographic CT images according to claim 1, characterized in that: The level set evolution model is established to make the curve evolve accurately to the edge of normal pulmonary artery tissue, and the pulmonary embolism lesion tissue is obtained after removing the normal pulmonary artery tissue, including: Based on the pulmonary artery data in the pulmonary artery and vein separation results, the mean and variance information of the pulmonary artery is counted and a pulmonary artery grayscale information model is established; According to the pulmonary artery grayscale information model, a length constraint item, a grayscale constraint item and an image information item are obtained; The knot length constraint term and image information term are used to establish the level set evolution model, and the embolized tissue is obtained through the level set curve evolution.

Images