A method for extracting blood vessels from head and neck CTA

Abstract

The present invention provides a head and neck CTA blood vessel extraction method, comprising the following steps: acquiring a head and neck CT image and a head and neck CTA image, aligning the head and neck CT image and the head and neck CTA image to obtain a first mask image of the blood vessel; extracting feature points in the first mask image of the blood vessel and classifying them through a deep learning model to obtain arterial point information; cropping the first mask image of the blood vessel to obtain a second mask image of the neck area; extracting the carotid artery and vertebral artery from the second mask image based on the arterial point information; and extending the carotid artery and the vertebral artery upward and downward, respectively, to achieve the extraction of four arteries. The present invention combines traditional algorithms with deep learning algorithms, and uses the arterial information obtained by the model to perform artery priority processing, which greatly reduces the influence of residue and venous reflux, obtains accurate centerline paths of the internal carotid artery and vertebral artery, reduces manual correction intervention, and increases clinical convenience and accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a head and neck CTA blood vessel extraction method. Background Art

[0002] CTA (Computed Tomography Angiography) is a technique that helps analyze and diagnose vascular diseases and conditions by injecting a contrast agent into the body's blood vessels and reconstructing CT images of the contrast agent.

[0003] Commonly used to examine cerebrovascular diseases such as vascular blockage, malformations, and soft plaques, CTA offers high spatial resolution. With technological advancements, head and neck CTA images are increasingly similar to DSA images, and in many cases, they can replace DSA. CTA vascular analysis is a key analytical tool for diagnosing head and neck vascular diseases. Automatically naming major blood vessels in the head and neck can improve diagnostic efficiency.

[0004] The accuracy of arterial segmentation determines the quality of CTA. During head and neck CTA scans, due to variations in heart rate and respiration, not only arteries but also numerous veins are present in the images. This results in low accuracy in automatic vessel naming (accurately identifying the left / right carotid artery and left / right vertebral artery), often requiring manual correction.

[0005] The current traditional method for naming vessels in head and neck CTAs typically involves first registering CT and CTA images, then performing a subtraction operation to obtain a bone-free vascular image. The corresponding centerline path is then manually selected by clicking on the vessel's initial and final points. This algorithm has the following disadvantages: 1. It requires manual intervention; 2. If the contrast agent level is too low or there is reflux, centering errors can occur, significantly reducing accuracy; 3. Manual modification is complex and inefficient.

[0006] Furthermore, standalone AI algorithms present another challenge: training set labeling and accuracy. The complex, numerous, and small-radius nature of the head and neck blood vessels, coupled with the need to consider lesions, makes labeling extremely challenging. This requires the expertise of experienced physicians, and a complete arterial labeling of a head and neck CTA can take several hours. AI algorithms, however, require vast amounts of data and are extremely costly. Current accuracy rates are approximately 92-93%, still far from ideal.

[0007] Current CTA algorithms have the following shortcomings: (1) Using purely traditional algorithms, there are interferences such as bone debris or venous reflux, low accuracy in vascular naming, and manual intervention is often required. (2) Using purely AI algorithms, there are difficulties in labeling data sets and insufficient data. Summary of the Invention

[0008] In order to overcome the above technical deficiencies, the present invention aims to provide a head and neck CTA blood vessel extraction method with higher accuracy.

[0009] The present invention discloses a head and neck CTA blood vessel extraction method, comprising the following steps: acquiring a head and neck CT image and a head and neck CTA image, and obtaining a first mask image of the blood vessels after registering the head and neck CT image and the head and neck CTA image; extracting feature points in the first mask image of the blood vessels and classifying them through a deep learning model to obtain arterial point information; cropping the first mask image of the blood vessels to obtain a second mask image of the neck area; extracting the carotid artery and the vertebral artery from the second mask image based on the arterial point information; and extending the carotid artery and the vertebral artery upward and downward, respectively, to achieve the extraction of four arteries.

[0010] Preferably, the extracting of feature points from the first mask image of the blood vessel and classifying them through a deep learning model to obtain arterial point information includes: slicing the head and neck CTA image in a preset range in the three directions of x, y, and z; and inputting the three slices into the deep learning model for classification to obtain arterial point information.

[0011] Preferably, the extracting of feature points from the first mask image of the blood vessel and classifying them through a deep learning model to obtain arterial point information also includes: binarizing the first mask image of the blood vessel; performing a three-dimensional distance transformation on the binarized first mask image of the blood vessel to obtain the shortest distance value from each pixel to a preset target pixel; extracting the local maximum values of the shortest distance value in several preset local areas, and performing dilution sampling on the local maximum values; and slicing the head and neck CTA image in a preset range in the three directions of x, y, and z with the local maximum value as the center.

[0012] Preferably, the extraction of feature points in the first mask image of the blood vessel and classification through a deep learning model, before obtaining arterial point information, also includes: obtaining a preset number of head and neck CTA data, and labeling the arteries, veins, and bones in the head and neck CTA data; dividing the head and neck CTA data into a training set, a test set, and a validation set; performing classification training, testing, and validation on the head and neck CTA data through a classification network to distinguish between arteries and non-arteries, thereby obtaining the deep learning model; the classification network includes ResNet, DenseNet, and VGG.

[0013] Preferably, the extending the carotid artery and the vertebral artery upward and downward respectively includes: extending the carotid artery and the vertebral artery upward to the head and downward to the heart respectively.

[0014] Preferably, when the carotid artery and vertebral artery are extracted from the second mask image based on the arterial point information, the rule includes: the CT values of the selected carotid artery and vertebral artery are greater than 200Hu.

[0015] Preferably, when extracting the carotid artery and vertebral artery from the second mask image based on the arterial point information, a rule includes: the radius of the carotid artery is greater than the radius of the vertebral artery.

[0016] Preferably, the extracting of the carotid artery and the vertebral artery from the second mask image based on the arterial point information includes: distinguishing the left and right positions and the center position of the blood vessels.

[0017] Preferably, obtaining the first mask image of the blood vessel after registering the CT data and the CTA data includes: registering the CT data and the CTA data and then performing subtraction to obtain the first mask image of the blood vessel.

[0018] Preferably, extracting the carotid artery and vertebral artery from the second mask image based on the arterial point information includes: removing the carotid artery from the second mask image to obtain a third mask image, and extracting the vertebral artery from the third mask image.

[0019] Compared with the existing technology, the above technical solution has the following beneficial effects:

[0020] 1. This invention combines traditional algorithms with deep learning algorithms, and uses the arterial information obtained by the model to perform arterial priority processing, which greatly reduces the impact of residue and venous return, obtains accurate centerline paths of the internal carotid artery and vertebral artery, reduces manual correction intervention, and increases the convenience and accuracy of clinical application; and in the case of low contrast agent concentration, the accuracy is also significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Flowchart of the head and neck CTA blood vessel extraction method provided by the present invention;

[0022] Figure 2 A flow chart of the present invention for classifying feature points through a deep learning model to obtain arterial point information;

[0023] Figure 3 The original blood vessel image during the deep learning model extraction process provided by the present invention;

[0024] Figure 4 These are the feature points extracted by the deep learning model provided by the present invention, where the green feature points are vein feature points and the red feature points are artery feature points;

[0025] Figure 5The set of arterial feature points finally extracted by the deep learning model in the deep learning model extraction process provided by the present invention;

[0026] Figure 6 This is the extraction result of the left carotid artery using the traditional algorithm;

[0027] Figure 7 The extraction result of the left artery provided by the present invention;

[0028] Figure 8 This is the extraction result of the right artery using the traditional algorithm;

[0029] Figure 9 This is the extraction result of the right artery provided by the present invention. DETAILED DESCRIPTION

[0030] The advantages of the present invention are further described below with reference to the accompanying drawings and specific embodiments.

[0031] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

[0032] The terms used in this disclosure are for the purpose of describing specific embodiments only and are not intended to limit the disclosure. As used in this disclosure and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0033] It should be understood that although the terms first, second, third, etc. may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining."

[0034] In the description of the present invention, it should be understood that the terms "longitudinal", "transverse", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention.

[0035] In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the internal communication between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.

[0036] In the following description, the suffixes such as "module", "component" or "unit" used to represent elements are only used to facilitate the description of the present invention and have no specific meaning. Therefore, "module" and "component" can be used interchangeably.

[0037] See attached Figure 1 The present invention discloses a head and neck CTA blood vessel extraction method, comprising the following steps:

[0038] S100, acquiring a head and neck CT image and a head and neck CTA image, and obtaining a first mask image of blood vessels after registering the head and neck CT image and the head and neck CTA image;

[0039] S200, extracting feature points from the first mask image of the blood vessel and classifying them through a deep learning model to obtain arterial point information;

[0040] S300, cropping the first mask image of the blood vessels to obtain a second mask image of the neck region;

[0041] S400, extracting the carotid artery from the second mask image based on the artery point information;

[0042] S500, after removing the carotid artery from the second mask image, obtaining a third mask image, and extracting the vertebral artery from the third mask image;

[0043] S600: Extend the carotid artery and vertebral artery upward and downward respectively to extract four arteries.

[0044] It should be noted that the above steps S100 to S500 do not represent a strict order of steps, but are only for reflecting the differences between the steps. For example, step S200 may also be after step S300.

[0045] In the above-mentioned head and neck CTA vessel extraction method, step S200 is an AI deep learning method, and step S300 is a traditional extraction method. This invention integrates the traditional algorithm with the AI algorithm, increasing the information dimension of the traditional algorithm and significantly improving the accuracy of automatic vessel naming.

[0046] The data registration in step S100 refers to the spatial alignment of multiple CTA scan images for accurate comparison and analysis. Registration can eliminate the position and posture differences between different scans, so that head and neck CTA images at different time points or different patients can be intuitively compared. CTA data registration generally includes the following steps. Image preprocessing: Preprocess the head and neck CTA images, including operations such as denoising, enhancement, and image smoothing. These steps help improve the accuracy and stability of registration. Feature extraction: Extract feature points or feature descriptors for registration from the head and neck CTA images. Common features can include vascular structures, key points, edges, etc. Feature matching: Match the features extracted from different head and neck CTA images to find the corresponding correspondence. The matching algorithm can use methods based on similarity metrics, such as nearest neighbor matching, optimal matching, etc. Transformation estimation: Based on the results of feature matching, estimate the spatial transformation relationship between images, such as translation, rotation, scaling, etc. Common transformation models include rigid transformation and affine transformation. Transformation Application: The estimated transformation is applied to the head and neck CTA image to align it to a reference image or a common coordinate system. This can be achieved through pixel-level image transformation using interpolation methods. Result Evaluation: The quality of the registered image is evaluated, including alignment accuracy, overlap, and visualization of the registration results. The registration process of the present invention utilizes one or more of the above steps, which are not limited here and can be adjusted based on actual needs.

[0047] Furthermore, after registering the CT and CTA data, subtraction is performed to obtain a first mask image of the blood vessels. Data subtraction is an image processing technique used to reduce artifacts or excessive contrast caused by vascular contrast agents in head and neck CTA images. Artifacts and excessive contrast can interfere with image visualization and analysis, so subtraction techniques can improve the quality and readability of head and neck CTA images. CTA data subtraction typically includes the following methods: Preprocessing: Preprocessing of head and neck CTA images, including denoising and smoothing. Denoising reduces noise interference in the image, and smoothing eliminates high-frequency noise. Vessel segmentation: Using image segmentation algorithms, vascular regions are extracted from head and neck CTA images. Vessel segmentation helps the subtraction algorithm process vascular regions more accurately and minimizes the impact on surrounding tissue. Modeling: A vascular contrast agent transmission model is developed to simulate light propagation within blood vessels. This helps estimate light attenuation and scattering within blood vessels. Compensation correction: Compensation correction is performed on head and neck CTA images based on the vascular contrast agent transmission model. This corrects the contrast of vascular regions and reduces artifacts. Post-processing: Further processing is performed on the corrected images, such as enhancing the visualization of vascular structures and removing residual artifacts. This improves image clarity and readability. CTA data subtraction technology can improve the quality of head and neck CTA images, reducing the effects of artifacts and excessive contrast, providing a more accurate and clear display of vascular structures, and helping physicians more accurately observe and analyze vascular lesions, hemodynamics, and other information.

[0048] For further information, see the attached Figure 2-5 Extracting feature points from the first mask image of the blood vessels in step S200 and classifying them through a deep learning model to obtain arterial point information specifically includes:

[0049] S201, performing binarization processing on the first mask image of the blood vessel;

[0050] S202, performing a three-dimensional distance transform on the binary first mask image of the blood vessel to obtain the shortest distance value from each pixel to a preset target pixel;

[0051] S203, extracting local maximum values of the shortest distance values in several preset local areas, and performing dilution sampling on the local maximum values;

[0052] S204, slicing the head and neck CTA image in a preset range in the x, y, and z directions, with the local maximum as the center;

[0053] S205: Input the three slices into a deep learning model for classification to obtain arterial point information.

[0054] The three-dimensional distance transform in step S202 is an image processing technique used to calculate the distance from each pixel in an image to a specified target. It can be used to analyze distance relationships between different regions in an image, such as in applications such as edge detection, morphological operations, and image segmentation. The basic algorithm for the three-dimensional distance transform includes the following steps: Initialization: Initialize the distance values of all pixels to infinity, except for the target pixel, which is set to 0. Pixel scanning: Starting from the target pixel, traverse each pixel in the image in a specific order. Distance update: For each pixel, calculate the distance values of the pixels in its neighborhood and update the current pixel with the minimum distance value plus 1. This gradually expands the distance values until the entire image is covered. Iterative scanning: Repeat the pixel scanning and distance update steps until the distance values for all pixels stabilize. Through the three-dimensional distance transform, the shortest distance value from each pixel to the target pixel is obtained. This distance representation forms a distance field in the image, which can be used to analyze and manipulate image structure and features. By calculating the distance relationship between pixels, the three-dimensional distance transform can extract geometric information from the image, helping to understand the image's structure and morphological features.

[0055] The deep learning model of the present invention adopts a simple binary classification algorithm model, which divides the input samples into one of two categories. Compared with the model structure that directly relies entirely on the deep learning model, it is relatively simple. For example - Logistic Regression: Logistic regression is a linear classification model that predicts the category probability of the sample by fitting a logical function. It is suitable for linearly separable or approximately linearly separable problems. Decision Tree: Decision tree is a classification algorithm based on a tree structure that divides the input samples into different categories through a series of splitting rules. It has the characteristics of good interpretability and easy understanding. Random Forest: Random Forest is an integrated learning method that performs classification by combining multiple decision trees. It performs modeling by randomly selecting features and sample subsets, which can improve the generalization ability and stability of the model. Support Vector Machine (SVM): Support Vector Machine is a classification algorithm based on statistical learning theory that performs classification by finding the optimal hyperplane in the feature space. It can handle high-dimensional data and nonlinear classification problems. K-Nearest Neighbors (KNN): This is an instance-based classification algorithm that measures the distance between an input sample and training samples. It determines the sample's category by voting based on the categories of its K nearest neighbors. Support Vector Machine (SVM): This is a classification algorithm based on statistical learning theory that classifies by finding the optimal hyperplane in feature space. It can handle high-dimensional data and nonlinear classification problems.

[0056] The method for constructing a deep learning model of the present invention is as follows: obtaining a preset amount of head and neck CTA data, and labeling the arteries, veins, and bones in the head and neck CTA data; dividing the head and neck CTA data into a training set, a test set, and a validation set, and performing classification training, testing, and validation on the head and neck CTA data through a classification network to distinguish between arteries and non-arteries, thereby obtaining a deep learning model.

[0057] The deep learning models obtained are applicable to different head and neck CTA data. The classification networks used here include but are not limited to ResNet, DenseNet, and VGG.

[0058] The deep learning model training of the present invention does not require a lot of AI data on the head and neck arteries. Preferably, 50 cases are enough. Compared with the existing technical solutions that rely entirely on deep learning models to extract blood vessels, the amount of calculation will be much less.

[0059] Preferably, extending the carotid artery upward and the vertebral artery downward respectively means extending the carotid artery upward to the head and extending the vertebral artery downward to the heart (if the heart is visible) respectively.

[0060] Preferably, when extracting the carotid artery and vertebral artery from the second mask image based on the arterial point information, it is necessary to use a vessel extraction rule based on CT value, vessel radius, vessel length, position information, etc., specifically:

[0061] (1) CT value: The CT value of the selected carotid artery and vertebral artery is greater than 200Hu.

[0062] (2) Vascular radius: The radius of the carotid artery is larger than that of the vertebral artery.

[0063] (3) Blood vessel length: When screening blood vessels, those with longer lengths are given priority.

[0064] (4) Position information: distinguish the left and right positions and the center position of the blood vessels.

[0065] Compared with traditional methods or AI methods, the present invention makes full use of the position, length, radius, CT value and other information involved in traditional methods, and integrates the classification information in AI methods to achieve artery classification, which greatly improves the accuracy of automatic vascular naming. Figure 6-9 , Figure 6 In the traditional algorithm, the left carotid artery is incorrectly extracted due to venous reflux and non-arterial residue. However, the present invention eliminates the influence of residue and reflux and obtains the correct result. Figure 8 In the traditional algorithm, the right carotid artery and the right vertebral artery cannot be extracted due to the low concentration of contrast agent. However, the present invention obtains the arterial feature point information through a deep learning model and successfully extracts the right artery.

[0066] It should be noted that the embodiments of the present invention have better practicability and do not impose any form of limitation on the present invention. Any technician familiar with the field may use the technical content disclosed above to change or modify it into an equivalent effective embodiment. However, any modification or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A head and neck CTA blood vessel extraction method, characterized in that: The steps include: Acquiring a head and neck CT image and a head and neck CTA image, and registering the head and neck CT image with the head and neck CTA image to obtain a first mask image of the blood vessels; Extracting feature points from the first mask image of the blood vessel and classifying them through a deep learning model to obtain arterial point information; cropping the first mask image of the blood vessel to obtain a second mask image of the neck region; extracting the carotid artery and the vertebral artery from the second mask image based on the artery point information; Extending the carotid artery and the vertebral artery upward and downward respectively to achieve the extraction of four arteries; Extracting feature points from the first mask image of the blood vessel and classifying them through a deep learning model to obtain arterial point information includes: Slicing the head and neck CTA image in a preset range in the x, y, and z directions respectively; Inputting the three slices into the deep learning model for classification to obtain arterial point information; The extracting feature points from the first mask image of the blood vessel and classifying them through a deep learning model to obtain arterial point information further includes: performing a binarization process on the first mask image of the blood vessel; Performing a three-dimensional distance transformation on the binary first mask image of the blood vessel to obtain the shortest distance value from each pixel to a preset target pixel; Extracting local maximum values of the shortest distance value in a plurality of preset local areas, and performing dilution sampling on the local maximum values; Slicing the head and neck CTA image in a preset range in the x, y, and z directions with the local maximum as the center; The first mask image of the blood vessel obtained by registering the CT data and the CTA data includes: The CT data and the CTA data are registered and then subtracted to obtain a first mask image of the blood vessel.

2. The head and neck CTA blood vessel extraction method according to claim 1, characterized in that: Before extracting the feature points in the first mask image of the blood vessel and classifying them through a deep learning model to obtain arterial point information, the method further includes: Acquire a preset amount of head and neck CTA data, and label arteries, veins, and bones in the head and neck CTA data; The head and neck CTA data are divided into a training set, a test set, and a validation set; performing classification training, testing, and validation on the head and neck CTA data using a classification network to distinguish arterial and non-arterial data, thereby obtaining the deep learning model; The classification network includes ResNet, DenseNet, and VGG.

3. The head and neck CTA blood vessel extraction method according to claim 1, characterized in that: The carotid artery and the vertebral artery are extended upward and downward respectively, comprising: The carotid artery and the vertebral artery extend upward to the head and downward to the heart respectively.

4. The head and neck CTA blood vessel extraction method according to claim 1, characterized in that: When extracting the carotid artery and vertebral artery from the second mask image based on the artery point information, the rules include: The CT values of the selected carotid artery and the selected vertebral artery are greater than 200Hu.

5. The head and neck CTA blood vessel extraction method according to claim 1, characterized in that: When extracting the carotid artery and vertebral artery from the second mask image based on the artery point information, the rules include: The radius of the carotid artery is greater than the radius of the vertebral artery.

6. The head and neck CTA blood vessel extraction method according to claim 1, characterized in that: The extracting of the carotid artery and the vertebral artery from the second mask image based on the artery point information includes: Distinguish the left, right, and center positions of blood vessels.

7. The head and neck CTA blood vessel extraction method according to claim 1, characterized in that: The extracting of the carotid artery and the vertebral artery from the second mask image based on the artery point information includes: After removing the carotid artery from the second mask image, a third mask image is obtained, and the vertebral artery is extracted from the third mask image.

Images