CTA Image-Based Intracranial Vascular Central Path Extraction Method and Its Device

Abstract

The present invention discloses a method and device for extracting the central path of intracranial blood vessels based on CTA images. The method includes: determining the root nodes of the intracranial blood vessel tree according to the root node prediction model and the CTA image to be processed; wherein, the root nodes of the intracranial blood vessel tree include the starting points of the left and right internal carotid arteries and the starting points of the left and right vertebral arteries; determining the tracking starting points of the central path of the intracranial blood vessels according to the tracking starting point prediction model and the CTA image to be processed; wherein, the tracking starting points include multiple central points of the intracranial blood vessel tree; traversing the tracking starting points, and for each root node, respectively extracting the intracranial blood vessel tree of the left and right internal carotid arteries and / or the intracranial blood vessel tree of the left and right vertebral arteries. Through the determination of the root nodes of the intracranial blood vessel tree and the tracking starting points of the blood vessel central path, the formulation of the blood vessel central path tracking strategy, and the determination of the end of the blood vessel central path tracking, the efficient and accurate extraction of the intracranial blood vessel central path can be realized, and the situation of excessive manual participation can be avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a method and a device for extracting the central path of intracranial blood vessels based on CTA images. Background Art

[0002] Intracranial vascular disease is a type of disease that causes brain tissue damage due to blood circulation disorders in intracranial blood vessels. It has a high disability and mortality rate, and early screening and prognosis and rehabilitation are crucial. Computed Tomography Angiography (CTA), as a commonly used examination technology for intracranial vascular diseases, has fast imaging speed and high resolution. However, traditional manual film reading is inefficient. In most cases, doctors can only "conceive" the spatial geometric relationship between the lesion and its surrounding tissues from multiple two-dimensional images. With the increase in the number of CTA examinations, it not only increases the workload of doctors, but also reduces the efficiency of clinical diagnosis. Through the three-dimensional reconstruction of blood vessels, the details of the major arteries of the brain can be displayed, assisting doctors in early detection of hidden lesions of patients' intracranial blood vessels. Therefore, automatic three-dimensional reconstruction of intracranial blood vessels based on CTA images has important clinical value.

[0003] At present, intracranial vascular segmentation based on CTA images is still very challenging. Due to the tortuous structure and numerous branches of intracranial blood vessels, it is difficult to accurately distinguish between blood vessels and artifacts. The segmented blood vessels are prone to interruption and are easily affected by other tissues. For example, the CT values ​​of blood vessels at the edge of the skull are highly similar to the CT values ​​of the skull, resulting in adhesion of extravascular tissues.

[0004] Intracranial blood vessels generally present a tree-like structure. The algorithm based on central path extraction can connect all the extracted central paths to form a vascular tree, which is used in the three-dimensional reconstruction of intracranial blood vessels. Existing technologies rely on the professional knowledge and long-term experience of researchers to select or construct potentially useful features, and require manual determination of the tracking starting point. More manual participation may result in low work efficiency and affect the accuracy of the extracted central path. Summary of the invention

[0005] 1. Purpose of the invention

[0006] The purpose of the present invention is to provide a method and device for extracting the central path of intracranial blood vessels based on CTA images, which can achieve efficient and accurate extraction of the central path of intracranial blood vessels and avoid situations where much manual participation is required.

[0007] (II) Technical solution

[0008] The first aspect of the present invention provides a method for extracting the central path of intracranial blood vessels based on CTA images, including: determining the root nodes of the intracranial blood vessel tree according to the root node prediction model and the CTA image to be processed; wherein, the root nodes of the intracranial blood vessel tree include the starting points of the left and right internal carotid arteries and the starting points of the left and right vertebral arteries; determining the tracking starting points of the central path of the intracranial blood vessels according to the tracking starting point prediction model and the CTA image to be processed; wherein, the tracking starting points include multiple central points of the intracranial blood vessel tree; traversing the tracking starting points, and for each root node, respectively extracting the intracranial blood vessel trees of the left and right internal carotid arteries and / or the intracranial blood vessel trees of the left and right vertebral arteries.

[0009] Further, the step of determining the root nodes of the intracranial blood vessel tree according to the root node prediction model and the CTA image to be processed includes: the root node prediction model learns the proximity features between all points in the training samples and the root nodes to obtain a trained root node prediction model; according to the trained root node prediction model and the CTA image to be processed, obtaining the proximity values of all points in the CTA image to be processed; sorting the proximity values of all points in the CTA image to be processed, and selecting candidate root nodes according to the set first candidate quantity; sequentially traversing the candidate root nodes, and screening out the root nodes that meet the preset conditions.

[0010] Further, the step of sequentially traversing the candidate root nodes and screening out the root nodes that meet the preset conditions includes: taking the point with the maximum proximity value among the candidate root nodes as the first root node; identifying the second root node on the other side according to the first root node; wherein, the second root node is at a preset first distance from the first root node; identifying the third root node and the fourth root node according to the first root node and the second root node; wherein, both the third root node and the fourth root node are at a preset first distance from one of the first root node and the second root node, and are at a preset second distance from the other of the first root node and the second root node; the third root node and the fourth root node are at a preset first distance from each other; calculating the coordinate sum of each root node according to the coordinates of the first root node, the second root node, the third root node and the fourth root node, and sorting according to the size of the coordinate sum to determine whether each root node corresponds to the starting point of the left internal carotid artery, the starting point of the right internal carotid artery, the starting point of the left vertebral artery or the starting point of the right vertebral artery.

[0011] Further, determining the tracking start point of the intracranial vascular central path according to the tracking start point prediction model and the CTA image to be processed includes: the tracking start point prediction model learning the proximity features between all points in the training samples and each point on the central path to obtain a trained tracking start point prediction model; obtaining the proximity values of all points in the CTA image to be processed according to the trained tracking start point prediction model and the CTA image to be processed; sorting the proximity values of all points in the CTA image to be processed, and selecting candidate tracking start points according to the set second candidate quantity; and screening out multiple tracking start points according to the candidate tracking start points.

[0012] Further, traversing the tracking start points, and respectively extracting the intracranial vascular trees of the left and right internal carotid arteries and / or the intracranial vascular trees of the left and right vertebral arteries for each root node includes: determining the tracking strategy to be executed for the vascular branches of each tracking start point according to the vascular radius information of each central point on the central path and the direction information from each current central point to the next central point in the training samples of the central path tracking model.

[0013] Further, determining the tracking strategy to be executed for the vascular branches of each tracking start point includes: if the vascular branch of the current tracking start point does not belong to the intracranial vascular tree, and the vascular branch of the current tracking start point is in forward tracking, when the vascular branch of the current tracking start point approaches the extracted intracranial vascular tree, determining the approaching intracranial vascular tree as the target intracranial vascular tree; or, if the vascular branch of the current tracking start point does not belong to the intracranial vascular tree, and the vascular branch of the current tracking start point is in reverse tracking, when the vascular branch of the current tracking start point approaches the extracted vascular tree, determining the approaching intracranial vascular tree as the target intracranial vascular tree; updating and storing the coordinates of the root node to which the target intracranial vascular tree belongs, and determining the end of the tracking of the vascular branch of the current tracking start point.

[0014] Further, determining the tracking strategy to be executed for the vascular branches of each tracking start point further includes: if the vascular branch of the current tracking start point belongs to the intracranial vascular tree, when the vascular branch of the current tracking start point approaches the extracted vascular tree, determining whether the approaching intracranial vascular tree and the intracranial vascular tree to which it belongs are the same root node; if they are the same root node, determining that the vascular branch of the current tracking start point is self-intersecting, and abandoning the tracking of the vascular branch of the current tracking start point; if they are not the same root node, continuing the tracking until the tracking of the vascular branch of the current tracking start point is determined to end; wherein, the completion of the tracking of the vascular branch of the current tracking start point is not affected by the currently approaching extracted intracranial vascular tree.

[0015] Further, the method for extracting the central path of intracranial blood vessels based on CTA images further includes: formulating a processing strategy for overlapping blood vessel segments where the central paths extracted for the overlapping blood vessel segments between different intracranial blood vessel trees do not coincide; wherein, formulating a processing strategy for overlapping blood vessel segments includes: traversing each intracranial blood vessel tree of each root node and traversing each blood vessel branch of the intracranial blood vessel tree; directly storing all the blood vessel branches of the intracranial blood vessel tree of the first root node; for the intracranial blood vessel trees of the second / third / fourth root nodes, traversing each of its blood vessel branches, screening out the first central points whose coordinates with the stored intracranial blood vessel tree are less than a preset distance, and marking the starting index of the first central points to be discarded; screening out the second central points whose coordinates with the stored intracranial blood vessel tree are greater than the preset distance, and marking the ending index of the second central points to be discarded; the number of marked and discarded points between the first central point and the second central point is less than a set threshold; or, the remaining central points after the second central point deviates from the direction of the first central point are less than a set threshold; then store the central points before the first central point deviates from the direction of the second central point; the number of marked and discarded points between the first central point and the second central point is greater than a set threshold; and the remaining central points after the second central point deviates from the direction of the first central point are greater than a set threshold; then store the central points before the first central point deviates from the direction of the second central point and store the central points after the second central point deviates from the direction of the first central point.

[0016] Further, the method for extracting the central path of intracranial blood vessels based on CTA images further includes: learning and training the neighborhood information of the end point of the central path and the neighborhood information of non-end points in the training samples according to the end point determination model, and obtaining a trained end point determination model; according to the trained end point determination model and the CTA image to be processed, if an end point is recognized, it is determined that the tracking of the intracranial blood vessel tree ends.

[0017] The second aspect of the present invention provides an apparatus for extracting the central path of intracranial blood vessels based on CTA images, including: a root node determination module, which is used to determine the root nodes of the intracranial blood vessel tree according to the root node prediction model and the CTA image to be processed; wherein, the root nodes of the intracranial blood vessel tree include the starting points of the left and right internal carotid arteries and the starting points of the left and right vertebral arteries; a tracking starting point determination module, which is used to determine the tracking starting point of the central path of the intracranial blood vessels according to the tracking starting point prediction model and the CTA image to be processed; wherein, the tracking starting point includes multiple central points of the intracranial blood vessel tree; an extraction module, which is used to traverse the tracking starting point, and for each root node, respectively extract the intracranial blood vessel trees of the left and right internal carotid arteries and / or the intracranial blood vessel trees of the left and right vertebral arteries.

[0018] (III) Beneficial effects

[0019] The above technical solutions of the present invention have the following beneficial technical effects:

[0020] 1. In the embodiments of the present invention, by determining the root node of the intracranial blood vessel tree and the starting point of the blood vessel center path tracking, formulating the blood vessel center path tracking strategy, and judging the end of the blood vessel center path tracking, the efficient and accurate extraction of the intracranial blood vessel center path can be realized, and the situation of excessive manual participation can be avoided; and the situation of blood vessel interruption caused by artifacts and the situation of adhesion of extravascular tissues can be better solved; and all the extracted center paths can be connected to form an intracranial blood vessel tree for three-dimensional reconstruction of intracranial blood vessels.

[0021] 2. Considering that the entire blood vessel tree may not be connected, the embodiments of the present invention set four root nodes for the intracranial blood vessel tree to independently extract the intracranial blood vessel tree connected to itself, which maximally guarantees the integrity of the intracranial blood vessel tree. Corresponding processing is also done for the situation where the center paths extracted from the overlapping blood vessel segments that may cause the intracranial blood vessel trees of different root nodes to overlap with each other are not completely coincident, which plays an important role in improving the accuracy of blood vessel center path extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flowchart of a method for extracting the intracranial blood vessel center path based on CTA images according to the first embodiment of the present invention;

[0023] Figure 2 is a flowchart of a method for extracting the intracranial blood vessel center path based on CTA images according to the second embodiment of the present invention;

[0024] Figure 3 Schematically shows the recognition results of the root node of the intracranial blood vessel tree and the tracking starting point;

[0025] Figure 4 Schematically shows the extraction results of the intracranial blood vessel tree center path;

[0026] Figure 5 is a schematic structural diagram of a device for extracting the intracranial blood vessel center path based on CTA images according to the third embodiment of the present invention;

[0027] Figure 6 is a schematic structural diagram of a device for extracting the intracranial blood vessel center path based on CTA images according to the fourth embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the specific embodiments and with reference to the accompanying drawings. It should be understood that these descriptions are exemplary and are not intended to limit the scope of the present invention. In addition, in the following descriptions, the descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present invention.

[0029] Intracranial blood vessels generally present a tree-like structure. The algorithm based on the extraction of the central path can connect all the extracted central paths to form a blood vessel tree, which is used in the three-dimensional reconstruction of intracranial blood vessels. This type of algorithm mainly includes three parts: the determination of the tracking starting point, the formulation of the tracking strategy, and the determination of the end of the tracking, which can better avoid the situation of blood vessel interruption caused by artifacts. For each blood vessel branch, the extraction process starts from the tracking starting point, tracks to its neighborhood, and classifies the points with consistent characteristics in the neighborhood into the path according to the tracking strategy until no new points can be classified into the path, that is, the tracking is determined to end. However, the traditional algorithms of this type rely on the professional knowledge and long-term experience of researchers to select or construct potentially useful features, and the tracking starting point needs to be determined manually. More manual participation may lead to low work efficiency and affect the accuracy of the extracted central path.

[0030] To solve these technical problems, the first aspect of the present invention provides a method for extracting the central path of intracranial blood vessels based on CTA images, as Figure 1 shown, specifically including the following steps:

[0031] Step S100, according to the root node prediction model and the CTA image to be processed, determine the root node of the intracranial blood vessel tree; wherein, the root node of the intracranial blood vessel tree includes the starting points of the left and right internal carotid arteries and the starting points of the left and right vertebral arteries;

[0032] Step S200, according to the tracking starting point prediction model and the CTA image to be processed, determine the tracking starting point of the central path of the intracranial blood vessels; wherein, the tracking starting point includes multiple central points of the intracranial blood vessel tree;

[0033] Step S300, traverse the tracking starting points, and for each root node, extract the intracranial blood vessel tree of the left and right internal carotid arteries and / or the intracranial blood vessel tree of the left and right vertebral arteries respectively.

[0034] Through deep learning technologies such as the root node prediction model and the tracking starting point prediction model, it does not rely on the professional knowledge and long-term experience of researchers to select or construct potentially useful features, and learns the most appropriate abstract features to a great extent from a certain amount of data, automatically identifying the root node of the intracranial blood vessel tree and the tracking starting point of the central path of the blood vessels; combined with formulating an appropriate tracking strategy for the central path of the blood vessels to achieve automatic tracking of the central path of the blood vessels and automatic determination of the end of the tracking, it can achieve efficient and accurate extraction of the central path of the intracranial blood vessels, and avoid the situation of more manual participation; and considering that the entire intracranial blood vessel tree may not be connected, four root nodes of the intracranial blood vessel tree are set, allowing them to independently extract their own connected intracranial blood vessel trees, which can ensure the integrity of the intracranial blood vessel tree to the greatest extent, and can better solve the situation of blood vessel interruption caused by artifacts and the situation of adhesion of extravascular tissues.

[0035] In some embodiments, in step S100, determining the root node of the intracranial vascular tree according to the root node prediction model and the CTA image to be processed includes:

[0036] Step S110, the root node prediction model learns the proximity features between all points in the training samples and the root node to obtain a trained root node prediction model;

[0037] Step S120, according to the trained root node prediction model and the CTA image to be processed, obtain the proximity values of all points in the CTA image to be processed;

[0038] Step S130, sort the proximity values of all points in the CTA image to be processed, and select candidate root nodes according to the set first candidate quantity; wherein, the first candidate quantity can be set to 500 - 1000 points.

[0039] Step S140, sequentially traverse the candidate root nodes and filter out the root nodes that meet the preset conditions.

[0040] In some embodiments, in step S140, sequentially traversing the candidate root nodes and filtering out the root nodes that meet the preset conditions specifically includes the following steps:

[0041] Step S141, take the point with the maximum proximity value among the candidate root nodes as the first root node;

[0042] Step S142, identify the second root node on the other side according to the first root node; wherein, the second root node is at a preset first distance from the first root node;

[0043] Step S143, identify the third root node and the fourth root node according to the first root node and the second root node; wherein, both the third root node and the fourth root node are at a preset first distance from one of the first root node and the second root node, and are at a preset second distance from the other of the first root node and the second root node; the third root node and the fourth root node are at a preset first distance from each other;

[0044] Step S144, calculate the coordinate sum of each root node according to the coordinates of the first root node, the second root node, the third root node, and the fourth root node, and sort them according to the size of the coordinate sum to determine whether each root node corresponds to the starting point of the left internal carotid artery, the starting point of the right internal carotid artery, the starting point of the left vertebral artery, or the starting point of the right vertebral artery.

[0045] The intracranial blood vessels can be divided into one connected blood vessel tree according to the actual situation (both the anterior communicating artery and the posterior communicating artery are present on both sides; only the anterior communicating artery, or only one side of the posterior communicating artery is absent), or two connected blood vessel trees (the anterior communicating artery is present and both sides of the posterior communicating artery are absent; the anterior communicating artery is absent and one side of the posterior communicating artery is absent), or three connected blood vessel trees (both the anterior communicating artery and the posterior communicating artery are absent on both sides). Considering the situation where the entire blood vessel tree may not be connected, the intracranial blood vessel tree can be set to include four root nodes, respectively at the starting positions of the left internal carotid artery (Left Internal Carotid Artery, LICA), the right internal carotid artery (Right Internal Carotid Artery, RICA), the left vertebral artery (Left Vertebral Artery, LVA), and the right vertebral artery (Right Vertebral Artery, RVA).

[0046] Training the root node prediction model: Using the CTA images with labeled root nodes as training samples and inputting them into the root node prediction model. By using deep learning technology, the root node prediction model learns the proximity features between all points in the training samples and the root nodes. The proximity threshold can be set to 16 mm, that is, the proximity value of points outside a distance of 16 mm from the center point is 0, and the proximity value of points within 16 mm decreases as the distance increases, thus obtaining a trained root node prediction model.

[0047] Predicting the root nodes: Using the CTA image to be processed as a test sample and inputting it into the trained root node prediction model to obtain the proximity values of all points in the CTA image to be processed. Sort the proximity values of all points in the CTA image to be processed, and select the top 800 points with the largest proximity values as candidate root nodes.

[0048] In an exemplary embodiment, the specific determination steps for the four root nodes corresponding to the starting points of the left internal carotid artery, the starting point of the right internal carotid artery, the starting point of the left vertebral artery, or the starting point of the right vertebral artery are as follows:

[0049] (1) Select the point with the largest proximity value among the candidate root nodes as the first root node, and search for the root node on the other side in sequence as the second root node; if the first root node is on the left, search for a point that meets the preset conditions on the right as the second root node, that is, the second root node is at a preset first distance from the first root node; the preset first distance can be set to 20 - 80 mm;

[0050] (2)Both the third and fourth root nodes found must satisfy being at a preset first distance of 20 - 80 mm from one of the first two root nodes and at a preset second distance from the other. The preset second distance can be set to 8 - 25 mm; and the distance between the third and fourth root nodes must satisfy a preset first distance of 20 - 80 mm, that is, find different types of root nodes on the same side.

[0051] (3)Calculate the coordinate sum of each root node based on the coordinates of the first, second, third, and fourth root nodes, that is, calculate the sum of the numerical values of the coordinates of the root nodes in the x, y, and z directions, and sort them according to the size of the coordinate sum. The size of the coordinate sum is sorted as LVA > LICA > RVA > RICA. Thus, four root nodes can be determined. The large dots shown in Figure 3 can be referred to, which represent the recognition results of the root nodes of the intracranial vascular tree.

[0052] In the embodiment of the present invention, considering that the entire intracranial vascular tree may not be connected, four root nodes of the intracranial vascular tree are set to independently extract the intracranial vascular tree connected to itself, which can ensure the integrity of the intracranial vascular tree to the greatest extent and can better solve the situation of vascular interruption caused by artifacts and the situation of adhesion of extravascular tissues.

[0053] In some embodiments, in step S200, determining the tracking starting point of the intracranial vascular central path according to the tracking starting point prediction model and the CTA image to be processed includes:

[0054] In step S210, the tracking starting point prediction model learns the proximity features between all points in the training samples and each center point of the central path to obtain a trained tracking starting point prediction model.

[0055] In step S220, according to the trained tracking starting point prediction model and the CTA image to be processed, obtain the proximity values of all points in the CTA image to be processed.

[0056] In step S230, sort the proximity values of all points in the CTA image to be processed, and select candidate tracking starting points according to the set second candidate quantity; among them, the second candidate quantity can be set to 1500 - 2000 center points.

[0057] In step S240, screen out multiple tracking starting points according to the candidate tracking starting points.

[0058] Training the tracking starting point prediction model: Use the CTA images with the extracted intracranial vascular tree as training samples and input them into the tracking starting point prediction model. Utilize deep learning technology to learn the proximity features between all points in the training samples and each point on the central path. The proximity threshold can be set to 4 mm, that is, the proximity value of points outside 4 mm from the center point is 0, and the proximity value of points within 4 mm decreases as the distance increases, thereby obtaining a trained tracking starting point prediction model.

[0059] Predicting the tracking starting point: Use the CTA images to be processed as test samples and input them into the trained tracking starting point prediction model to obtain the proximity values of all points in the CTA images to be processed. Sort the proximity values of all points in the CTA images to be processed, select the top 1600 points with the largest proximity values as candidate tracking starting points, and then use the Non-Maximum Suppression (NMS) algorithm to exclude points within 2 mm. Finally, retain approximately 400 points as tracking starting points among the candidate tracking starting points for forward tracking and backward tracking. The small dots shown in Figure 3 represent the recognition results of the intracranial vascular tree tracking starting points.

[0060] In some embodiments, the forward direction can be set as the direction with the highest prediction probability in the direction vector of the current tracking starting point; the backward direction can be set as the direction that satisfies the angle with the forward direction ≥ 90° and has the second highest prediction probability in the direction vector of the current tracking starting point.

[0061] In some embodiments, in step S300, when traversing the tracking starting points, for each root node, respectively extracting the intracranial vascular tree of the left and right internal carotid arteries and / or the intracranial vascular tree of the left and right vertebral arteries includes:

[0062] According to the central path tracking model, learn the blood vessel radius information (as the step size) of each central point on the central path in the training samples and the direction information from each current central point to the next central point to determine the tracking strategy for the blood vessel branches of each tracking starting point. Use the CTA images with the labeled central path as training samples and input them into the central path tracking model. Utilize deep learning technology to let the central path tracking model learn the blood vessel radius information (as the step size) of each central point on the central path in the training samples and the direction information from each current central point to the next central point, and a trained central path tracking model can be obtained. Use the CTA images to be processed as test samples and input them into the trained central path tracking model to determine the tracking strategy that can be executed for the blood vessel branches of each tracking starting point.

[0063] In some embodiments, determining the tracking strategy for the blood vessel branches of each tracking starting point includes:

[0064] If the blood vessel branch at the current tracking starting point does not belong to the intracranial blood vessel tree (i.e., the blood vessel branch at the current tracking starting point is stored as none), and the blood vessel branch at the current tracking starting point is in forward tracking, when the blood vessel branch at the current tracking starting point approaches the extracted intracranial blood vessel tree, then determine the approached intracranial blood vessel tree as the target intracranial blood vessel tree; or,

[0065] If the blood vessel branch at the current tracking starting point does not belong to the intracranial blood vessel tree, and the blood vessel branch at the current tracking starting point is in reverse tracking, that is, the forward tracking of the blood vessel branch at the current tracking starting point does not approach the extracted intracranial blood vessel tree, when the blood vessel branch at the current tracking starting point approaches the extracted blood vessel tree, then determine the approached intracranial blood vessel tree as the target intracranial blood vessel tree;

[0066] Update and store the coordinates of the root node to which the target intracranial blood vessel tree belongs, and determine the end of the tracking of the blood vessel branch at the current tracking starting point. Initially, the blood vessel branch at the current tracking starting point is default stored as none, and only when it is stored as none is it considered whether to update, that is, the blood vessel branch at the current tracking starting point has not been added to any intracranial blood vessel tree.

[0067] In some embodiments, the determining the tracking strategy performed on the blood vessel branch at each tracking starting point further includes:

[0068] If the blood vessel branch at the current tracking starting point belongs to the intracranial blood vessel tree, that is, the blood vessel branch at the current tracking starting point is in reverse tracking, and the forward tracking of the blood vessel branch at the current tracking starting point has approached the extracted intracranial blood vessel tree, when the blood vessel branch at the current tracking starting point approaches the extracted blood vessel tree, then determine whether the approached intracranial blood vessel tree and the intracranial blood vessel tree to which it belongs are the same root node to which they belong;

[0069] If they are the same root node to which they belong, then determine that the blood vessel branch at the current tracking starting point is self-intersecting, and discard the tracking of the blood vessel branch at the current tracking starting point;

[0070] If they are not the same root node to which they belong, then continue the tracking until it is determined that the tracking of the blood vessel branch at the current tracking starting point ends; among them, the completion of the tracking of the blood vessel branch at the current tracking starting point is not affected by the currently approached extracted intracranial blood vessel tree.

[0071] In some embodiments, a method for extracting the central path of intracranial blood vessels based on CTA images, as Figure 2 shown, may further include:

[0072] Step S400, determining the end of the tracking of the intracranial blood vessel tree. Among them, step S400 may include the following specific steps:

[0073] Step S410: According to the end - point determination model, learn the neighborhood information of the end - points of the central path and the non - end - point neighborhood information in the training samples, and obtain a trained end - point determination model. Input the CTA images with labeled end - points as training samples into the end - point determination model, and use deep - learning technology to enable the end - point determination model to learn the neighborhood information of the end - points of the central path and the non - end - point neighborhood information in the training samples, so as to obtain a trained end - point determination model.

[0074] Step S420: According to the trained end - point determination model and the CTA image to be processed, if an end - point is recognized, it is determined that the tracking of the intracranial vascular tree ends.

[0075] In some embodiments, when four root nodes independently extract the intracranial vascular trees connected to themselves, it may cause the central paths of the extracted vascular segments that overlap between the intracranial vascular trees of different root nodes not to completely coincide. The following processing steps can be adopted:

[0076] Step S500: For the non - overlapping central paths of the extracted vascular segments that overlap between different intracranial vascular trees, formulate a processing strategy for the overlapping vascular segments. Among them, formulating a processing strategy for the overlapping vascular segments includes:

[0077] Step S510: Traverse the intracranial vascular tree of each root node and each blood - vessel branch of the intracranial vascular tree.

[0078] Step S520: Directly store all the blood - vessel branches of the intracranial vascular tree of the first root node.

[0079] Step S530: For the intracranial vascular tree of the second / third / fourth root node, traverse each of its blood - vessel branches, screen out the first central points whose coordinates are less than a preset distance from the stored intracranial vascular tree, and mark the first central points as discard start indices; screen out the second central points whose coordinates are greater than the preset distance from the stored intracranial vascular tree, and mark the second central points as discard end indices. The preset distance can be set to 2 mm.

[0080] Step S540: If the number of marked discard points between the first central point and the second central point is less than the set threshold; or, the remaining central points after the second central point deviates from the first central point direction are less than the set threshold; then store the central points before the first central point deviates from the second central point direction; otherwise, if the conditions in Step S540 are not met, execute Step S550.

[0081] Step S550: If the number of marked discarded points between the first center point and the second center point is greater than a set threshold; and the remaining center points after the second center point deviates from the direction of the first center point are greater than the set threshold; then store the center points before the first center point deviates from the direction of the second center point, and store the center points after the second center point deviates from the direction of the first center point.

[0082] The set threshold can be 50 points, or the length of the marked discarded center path between the first center point and the second center point is about 15 mm; or the remaining center path length after the second center point deviates from the direction of the first center point is about 15 mm.

[0083] In an exemplary embodiment, all blood vessel branches of the intracranial vascular tree of the first root node are directly stored. For the intracranial vascular trees of other root nodes, each blood vessel branch is traversed. When a point close to the coordinates of the stored intracranial vascular tree (preset distance ≤ 2 mm) is encountered, mark the starting index start of the discard until a point not close to the coordinates of the vascular tree (preset distance > 2 mm) is encountered, and mark the ending index end of the discard. To avoid abrupt connection, 5 points before and after both the marked start and end can be retained;

[0084] If the number of discarded points is less than 50 points (about 15 mm), or the remaining points at the end after discarding are less than 50 points, then only store the points before start + 5, and vice versa, also store the points after end - 5. This can remove some miscellaneous points and solve the problem that the central paths extracted from overlapping blood vessel segments are not completely coincident, improving the accuracy of blood vessel central path extraction.

[0085] First, store the blood vessel branches processed by the above processing strategy for overlapping blood vessel segments in a temporary list, and then store them in the intracranial vascular tree coordinate list after all blood vessel branches of the intracranial vascular tree of the same root node are traversed. This can avoid comparing blood vessel branches of the same root node, that is, blood vessel branches belonging to the same root node need to be compared with blood vessel branches of different root nodes.

[0086] In the embodiments of the present invention, through the determination of the root node of the intracranial vascular tree and the starting point of blood vessel central path tracking, the formulation of the blood vessel central path tracking strategy, and the determination of the end of blood vessel central path tracking, and corresponding processing is also performed on the situation where the central paths extracted from overlapping blood vessel segments between intracranial vascular trees of different root nodes are not completely coincident, the central path extraction result of the intracranial vascular tree as shown in Figure 4 can be obtained, realizing efficient and accurate extraction of the intracranial blood vessel central path.

[0087] Based on the same inventive concept, the second aspect of the present invention provides an intracranial blood vessel central path extraction device based on CTA images, which is used to execute the above-mentioned intracranial blood vessel central path extraction method based on CTA images, as Figure 5As shown, the device includes:

[0088] A root node determination module 10, which is used to determine the root nodes of the intracranial vascular tree according to the root node prediction model and the CTA image to be processed; wherein, the root nodes of the intracranial vascular tree include the starting points of the left and right internal carotid arteries and the starting points of the left and right vertebral arteries;

[0089] A tracking starting point determination module 20, which is used to determine the tracking starting points of the central paths of the intracranial blood vessels according to the tracking starting point prediction model and the CTA image to be processed; wherein, the tracking starting points include multiple center points of the intracranial vascular tree;

[0090] An extraction module 30, which is used to traverse the tracking starting points, and for each root node, extract the intracranial vascular trees of the left and right internal carotid arteries and / or the intracranial vascular trees of the left and right vertebral arteries respectively.

[0091] The intracranial vascular central path extraction device based on the CTA image, as Figure 6 shown, may further include:

[0092] A determination of tracking end module 40, which is used to determine the end of the tracking of the intracranial vascular tree.

[0093] A processing module 50, which is used to formulate a processing strategy for the overlapping vascular segments when the central paths extracted from the overlapping vascular segments between different intracranial vascular trees do not coincide.

[0094] Wherein, in the embodiments of the present invention, the specific shapes and structures of the root node determination module 10, the tracking starting point determination module 20, the extraction module 30, the determination of tracking end module 40 and the processing module 50 are not limited. Those skilled in the art can set them arbitrarily according to their implemented functions and effects, and will not be elaborated here; in addition, the specific implementation processes and implementation effects of the operation steps implemented by the above-mentioned modules in the embodiments of the present invention are the same as those of steps S100 - S500 in the embodiments of the present invention. For details, reference can be made to the above description, and will not be elaborated here.

[0095] It should be understood that the above specific embodiments of the present invention are only used for exemplary illustration or explanation of the principle of the present invention, and do not constitute a limitation to the present invention. Therefore, any modifications, equivalent replacements, improvements, etc. made without departing from the spirit and scope of the present invention shall be included within the protection scope of the present invention. In addition, the appended claims of the present invention are intended to cover all changes and modification examples falling within the scope and boundaries of the appended claims, or equivalent forms of such scope and boundaries.

[0096] Those of ordinary skill in the art can understand that all or part of the steps in the methods of the above embodiments can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. When the program is executed, it includes the processes of the embodiments of the above various methods. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM for short), a random access memory (RAM for short), etc.

[0097] The steps in the method embodiments of the present invention can be adjusted, combined, and deleted according to actual needs. The modules in the system embodiments of the present invention can be combined, divided, and deleted according to actual needs.

Claims

1. A method for extracting the central path of intracranial blood vessels based on CTA images, characterized in that, Including: Determine the root nodes of the intracranial vascular tree according to the root node prediction model and the CTA image to be processed, specifically including: The root node prediction model learns the proximity features between all points in the training samples and the root nodes to obtain a trained root node prediction model; according to the trained root node prediction model and the CTA image to be processed, obtain the proximity values of all points in the CTA image to be processed; sort the proximity values of all points in the CTA image to be processed, and select candidate root nodes according to the set first candidate quantity; sequentially traverse the candidate root nodes and filter out the root nodes that meet the preset conditions; wherein, the root nodes of the intracranial vascular tree include the starting points of the left and right internal carotid arteries and the starting points of the left and right vertebral arteries; Determine the tracking starting points of the intracranial vascular central path according to the tracking starting point prediction model and the CTA image to be processed; wherein, the tracking starting points include multiple central points of the intracranial vascular tree; Traverse the tracking starting points, and for each root node, extract the intracranial vascular trees of the left and right internal carotid arteries and / or the intracranial vascular trees of the left and right vertebral arteries respectively.

2. The method according to claim 1, wherein The sequential traversal of the candidate root nodes to filter out the root nodes that meet the preset conditions includes: Take the point with the maximum proximity value among the candidate root nodes as the first root node; Identify the second root node on the other side according to the first root node; wherein, the second root node is at a preset first distance from the first root node; Identify the third root node and the fourth root node according to the first root node and the second root node; wherein, both the third root node and the fourth root node are at a preset first distance from one of the first root node and the second root node, and are both at a preset second distance from the other of the first root node and the second root node; the third root node and the fourth root node are at a preset first distance from each other; Calculate the coordinate sum of each root node according to the coordinates of the first root node, the second root node, the third root node and the fourth root node, and sort them according to the size of the coordinate sum to determine whether each root node corresponds to the starting point of the left internal carotid artery, the starting point of the right internal carotid artery, the starting point of the left vertebral artery or the starting point of the right vertebral artery.

3. The method according to claim 1, wherein The determining the tracking starting points of the intracranial vascular central path according to the tracking starting point prediction model and the CTA image to be processed includes: The tracking starting point prediction model learns the proximity features between all points in the training samples and each point of the central path to obtain a trained tracking starting point prediction model; According to the trained tracking starting point prediction model and the CTA image to be processed, obtain the proximity values of all points in the CTA image to be processed; Sort the proximity values of all points in the CTA image to be processed, and select candidate tracking starting points according to the set second candidate quantity; Filter out multiple tracking starting points according to the candidate tracking starting points.

4. The method according to claim 1, wherein The traversing the tracking starting points and, for each root node, respectively extracting the intracranial vascular trees of the left and right internal carotid arteries and / or the intracranial vascular trees of the left and right vertebral arteries includes: According to the central path tracking model, learn the blood vessel radius information of each central point on the central path in the training samples and the direction information from each current central point to the next central point, so as to determine the tracking strategy for the blood vessel branches starting from each tracking starting point.

5. The method according to claim 4, wherein The determination of the tracking strategy for the blood vessel branches starting from each tracking starting point includes: If the blood vessel branch starting from the current tracking starting point does not belong to the intracranial blood vessel tree, and the blood vessel branch starting from the current tracking starting point is in forward tracking, when the blood vessel branch starting from the current tracking starting point approaches the extracted intracranial blood vessel tree, then determine the approaching intracranial blood vessel tree as the target intracranial blood vessel tree; or, If the blood vessel branch starting from the current tracking starting point does not belong to the intracranial blood vessel tree, and the blood vessel branch starting from the current tracking starting point is in reverse tracking, when the blood vessel branch starting from the current tracking starting point approaches the extracted blood vessel tree, then determine the approaching intracranial blood vessel tree as the target intracranial blood vessel tree; Update and store the coordinates of the root node to which the target intracranial blood vessel tree belongs, and determine the end of the tracking of the blood vessel branch starting from the current tracking starting point.

6. The method according to claim 5, wherein The determination of the tracking strategy for the blood vessel branches starting from each tracking starting point further includes: If the blood vessel branch starting from the current tracking starting point belongs to the intracranial blood vessel tree, when the blood vessel branch starting from the current tracking starting point approaches the extracted blood vessel tree, then determine whether the approaching intracranial blood vessel tree and the intracranial blood vessel tree to which it belongs are the same root node; If they are the same root node, then determine that the blood vessel branch starting from the current tracking starting point is self-intersecting, and discard the tracking of the blood vessel branch starting from the current tracking starting point; If they are not the same root node, then continue the tracking until the tracking of the blood vessel branch starting from the current tracking starting point is determined to end; among them, the completion of the tracking of the blood vessel branch starting from the current tracking starting point is not affected by the currently approaching extracted intracranial blood vessel tree.

7. The method according to claim 1, characterized in that It further includes: Formulate a processing strategy for the overlapping blood vessel segments when the central paths extracted between different intracranial blood vessel trees do not coincide; among them, Formulating a processing strategy for the overlapping blood vessel segments includes: Traverse the intracranial blood vessel tree of each root node and each blood vessel branch of the intracranial blood vessel tree; Directly store all the blood vessel branches of the intracranial blood vessel tree of the first root node; For the intracranial blood vessel tree of the second / third / fourth root node, traverse each of its blood vessel branches, screen out the first central points whose coordinates are less than the preset distance from the stored intracranial blood vessel tree, and mark the first central points as the discarded starting indexes; screen out the second central points whose coordinates are greater than the preset distance from the stored intracranial blood vessel tree, and mark the second central points as the discarded ending indexes; If the number of the already marked discarded points between the first central point and the second central point is less than the set threshold; or, the remaining central points after the second central point deviates from the direction of the first central point are less than the set threshold; then store the central points before the first central point deviates from the direction of the second central point; The number of marked discarded points between the first center point and the second center point is greater than a set threshold; and the number of remaining center points after the second center point deviates from the direction of the first center point is greater than the set threshold; then the center points before the first center point deviates from the direction of the second center point are stored, and the center points after the second center point deviates from the direction of the first center point are stored.

8. The method according to claim 1, characterized in that, It further includes: According to the end point determination model, learning the neighborhood information of the end point of the central path and the neighborhood information of non-end points in the training sample to obtain a trained end point determination model; According to the trained end point determination model and the CTA image to be processed, if an end point is recognized, it is determined that the tracking of the intracranial vascular tree ends.

9. An intracranial vascular central path extraction device based on CTA images, characterized in that, It includes: A root node determination module, which is used to learn the proximity characteristics between all points and the root node in the training sample according to the root node prediction model to obtain a trained root node prediction model; according to the trained root node prediction model and the CTA image to be processed, obtain the proximity values of all points in the CTA image to be processed; sort the proximity values of all points in the CTA image to be processed, and select candidate root nodes according to the set first candidate quantity; sequentially traverse the candidate root nodes and screen out the root nodes that meet the preset conditions; among them, the root nodes of the intracranial vascular tree include the starting points of the left and right internal carotid arteries and the starting points of the left and right vertebral arteries; A tracking start point determination module, which is used to determine the tracking start point of the intracranial vascular central path according to the tracking start point prediction model and the CTA image to be processed; among them, the tracking start point includes multiple center points of the intracranial vascular tree; An extraction module, which is used to traverse the tracking start point, and for each root node, extract the intracranial vascular tree of the left and right internal carotid arteries and / or the intracranial vascular tree of the left and right vertebral arteries respectively.

Images