CN110338830A - A method for automatically extracting the central path of head and neck vessels in CTA images - Google Patents

Abstract

The present invention provides a kind of method for automatically extracting neck blood vessel center path in CTA image, this method is applicable in from the arch of aorta and scans to the CTA image data of calvarium, include the following steps: S1: according to blood vessel in the shape, gray scale and position feature in CTA image cross section, the beginning and end of arteria carotis and vertebral artery being automatically positioned respectively；S2: between the blood vessel beginning and end of step S1 positioning, the normalization vessel filter of the multi-scale gradient based on Raycasting is carried out to the region for meeting blood vessel gray feature, constructs vessel filter weight figure；S3: in the vessel filter weight figure of step S2 building, the extraction to arteria carotis and vertebral artery blood vessel center path is completed using Dijkstra optimal path extraction algorithm.Advantages of the present invention: realizing the automatic positioning of blood vessel, participates in manually without user；The center line of extraction is accurate, adjusts without user's later period；The method speed of service is fast, meets real-time demand.

Description

A method for automatically extracting the central path of head and neck vessels in CTA images

technical field

The invention relates to a method for automatically extracting the central path of head and neck blood vessels in CTA images. The invention belongs to the technical field of medical image processing.

Background technique

Cerebrovascular disease is one of the main diseases that threaten human health. In clinical practice, the three-dimensional structural information of blood vessels is obtained by analyzing the images after CTA imaging (ie, non-invasive vascular imaging examination), and extracting the central path of blood vessels. Doctors determine the degree of vascular disease. However, due to the complex distribution of blood vessels in the human body, blood vessels often pass through the bones, and both the bones and blood vessels show bright grayscale features in the images generated after CTA imaging, which makes the extraction of the central path of the blood vessels and the pathological changes. Analysis brings great difficulty.

At present, when performing blood vessel analysis on images generated after CTA images, subtraction technology is usually used to remove the bones in the CTA images to obtain the blood vessel area, and then manually define the start point and end point of the blood vessel to obtain the center of the blood vessel. path. The disadvantage of this method is: since the subtraction operation requires two scans of the patient (CT and CTA scans), the patient is easily displaced between the two scans, which will result in the loss of the vascular structure after subtraction. or rupture, which will affect the subsequent analysis of vascular lesions.

Using the Hessian-based vascular enhancement method to obtain the central path of the blood vessel, although the central path of the blood vessel can be directly extracted from the CTA image, for blood vessels with complex shapes and surrounding tissues, such as the vertebrae and the cranial segment of the internal carotid artery, usually The effect is not very ideal, and the doctor needs to manually define the starting point and ending point of the blood vessel, which brings a certain burden to the doctor's analysis work.

SUMMARY OF THE INVENTION

In view of the above reasons, the purpose of the present invention is to provide a method for automatically extracting the central paths of the head and neck vessels, ie, the vertebral artery vessels and the carotid vessels, in CTA images.

In order to achieve the above object, the present invention adopts the following technical scheme: a method for automatically extracting the central path of the head and neck blood vessels in the CTA image, the method is applicable to the CTA image data scanned from the aortic arch to the cranial top, and is characterized in that: it comprises the following steps:

S1: According to the shape, gray level and position characteristics of the blood vessels in the cross-section of the CTA image, the starting point and the end point of the carotid artery and the vertebral artery are automatically located respectively;

S2: Between the start point and the end point of the blood vessel located in step S1, perform multi-scale gradient normalization blood vessel filtering based on Raycasting on the region conforming to the grayscale characteristics of the blood vessel, and construct a blood vessel filtering weight map;

S3: In the blood vessel filtering weight map constructed in step S2, the Dijkstra optimal path extraction algorithm is used to complete the extraction of the central path of the carotid artery and the vertebral artery.

The advantages of the invention are as follows: the automatic positioning of the blood vessel is realized without the user's manual participation; the extracted centerline is accurate, and the user does not need to adjust later; the method runs quickly and meets the real-time requirement.

Description of drawings

Fig. 1 is the flow chart of the method for automatically extracting the central path of head and neck blood vessels according to the present invention;

Fig. 2 is a multi-scale gradient normalized blood vessel filtering diagram based on Raycasting (ray casting);

Figure 3A is the origin of the carotid artery and vertebral artery (green marking point) located after step S1.1.2;

Figure 3B is the end point of the vertebral artery (green marking point) located after step S1.2.1.2;

Figure 3C is the carotid artery end point (green marking point) located after step S1.2.2.2;

Fig. 3D is the right carotid artery central path extracted after step S3.2 and the corresponding CPR (surface reconstruction) result diagram;

Fig. 3E is the left carotid artery central path extracted after step S3.2 and its corresponding CPR (surface reconstruction) result diagram;

Fig. 3F is the right vertebral artery central path extracted after step S3.2 and the corresponding CPR (surface reconstruction) result diagram;

FIG. 3G is a graph of the central path of the left vertebral artery extracted after step S3.2 and its corresponding CPR (surface reconstruction) result.

Detailed ways

The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that various modifications can be made to the embodiments disclosed herein, therefore, the embodiments disclosed in the specification should not be regarded as limitations of the present invention, but only as examples of the embodiments, the purpose of which is to make the present invention The features of the invention are obvious.

As shown in FIG. 1 , the method for automatically extracting the central path of head and neck blood vessels (vertebral artery blood vessels and carotid arteries) in a CTA image provided by the present invention is applicable to the CTA image data scanned from the aortic arch to the top of the skull (described below as the CTA image that meets the requirements) data). The method provided by the present invention mainly includes three key steps:

S1: According to the shape, gray level and position characteristics of the blood vessels in the cross-section of the CTA image, the starting point and the end point of the carotid artery and the vertebral artery are automatically located respectively;

S2: between the start point and the end point of the blood vessel located in step S1, perform multi-scale gradient normalized blood vessel filtering based on Raycasting on the region that conforms to the grayscale characteristics of the blood vessel, and construct a blood vessel filtering weight map according to the calculated filter value;

S3: In the blood vessel filtering weight map constructed in step S2, the Dijkstra optimal path extraction algorithm is used to complete the extraction of the central path of the carotid artery and the vertebral artery.

The human vertebral artery originates from the subclavian artery, passes through the transverse foramen of the vertebrae, bends medially after the lateral mass of the atlas, passes through the foramen magnum into the cranial cavity, and joins with the contralateral vertebral artery to form the basilar artery at the lower border of the pons. Therefore, the present invention needs to locate the starting points of the left and right vertebral arteries, and simultaneously needs to locate the basilar artery points as the termination points of the left and right vertebral arteries.

The carotid arteries arise from the aortic arch and the subclavian arteries, and ascend upward to give rise to two branches, the internal carotid artery and the external carotid artery. The internal carotid artery extends upward through the ruptured hole into the cranial cavity, and enters the cavernous sinus through the inferior petrosal sinus, forming an S-shaped curve in the cavernous sinus, and then extends forward through the dura mater. Arteries that participate in the formation of the cerebral arterial ring. The present invention needs to locate the starting point of the left and right carotid arteries and the ending point of the internal carotid artery, respectively.

Step S1 of the present invention: According to the shape of the blood vessel in the cross-section of the CTA image, the gray level and the positional feature, the starting point and the end point of the carotid artery and the vertebral artery are respectively automatically positioned, and the specific steps are as follows:

S1.1 Automatically locate the origin of the vertebral artery and carotid artery

S1.1.1 Locating candidate slices for the origin of the carotid and vertebral arteries

First, perform binarization preprocessing on the CTA image data that meets the requirements to distinguish human tissue and background areas in the image; The direction of the top of the skull) is scanned layer by layer. When the area of a single hole is greater ^than 60mm2 and less ^than 1200mm2, the current layer is considered to be the upper edge of the lung parenchyma. At this time, this layer is used as a candidate layer for the origin of the vertebral artery and carotid artery, and stop scanning;

S1.1.2 Locate the origin of the vertebral and carotid vessels

In the candidate layer, all regions with gray value greater than 150HU are marked and extracted to obtain the candidate regions of the vertebral artery and carotid artery, and the candidate regions are screened using the following rules:

1) Calculate the center of gravity of the marked tissue, and remove the unmarked area of the center of gravity;

2) On the basis of step 1), calculate the adjacent side ratio of the minimum circumscribed rectangle of the remaining candidate area, and remove the area with the ratio greater than 3 or less than 1/3;

3) According to the size characteristics of the normal vertebral artery and the carotid artery, on the basis of steps 1) and 2), further remove the area of the candidate area greater than 60 mm ² and less than 3 mm ² ;

4) Calculate the grayscale mean and variance of the remaining candidate regions one by one, and take the center of gravity of the current candidate region as the seed point, take 2 times the variance as the growth step, and take the layer where the current candidate region is located as the benchmark for the upper and lower 3-layer image ranges Carry out three-dimensional region growth in the interior; the region grown from each layer is judged again according to the adjacent edge ratio in step 2), and the candidate regions that do not meet the conditions are removed;

5) The cross-section of the blood vessel appears as a round-like feature in the CTA image, and the distribution in the human body is approximately centrosymmetric. Therefore, the Circularity, Symmetry, and Bounding Box are calculated for the areas screened in step 4). Aspect ratio Ratio and area area Area, among which, the center symmetry indicates that the center of the human tissue after binarization (x direction) is used as the benchmark to find out whether there is a candidate area within 5mm of the opposite side, up, down, left, and right; if it can be found, calculate The distance between the center of gravity of all blood vessel candidate regions in the search range and the center of gravity of the current region in the y direction, and the corresponding minimum distance is taken as the center symmetry degree of the current blood vessel candidate area; if not found, the center symmetry degree of the current blood vessel candidate area is set to the side length of the entire image;

6) Sort the four measurement parameters calculated in step 5): the circularity and area are sorted from large to small, the aspect ratio of the bounding box is sorted from small to large according to the difference from 1, the center Symmetry is sorted from small to large, and the sorted order is used as the score of the measurement value of the area: the circularity score is expressed as S _circularity , the symmetry score is expressed as S _symmetry , the aspect ratio score is expressed as S _ratio , The area fraction is expressed as S _area ;

7) Define the similarity score of blood vessel section: Score=w1*S _circlarity +w2*S _symmetry +w3*S _ratio +w4*S _area , where w1, w2, w3, and w4 represent the circularity score of the candidate blood vessel area, Weights corresponding to symmetry score, aspect ratio score, and area score; the smaller the blood vessel similarity score, the more likely the area is to be a blood vessel area;

8) Sort the blood vessel similarity scores in order from small to large, and obtain the top four candidate regions; first, define the two larger areas as the starting areas of the left and right carotid arteries, and the other two are defined as The starting areas of the left and right vertebral arteries; then, the left and right carotid arteries and the left and right vertebral arteries are distinguished according to the obtained x-coordinates of the center of gravity of the starting areas of the carotid and vertebral arteries; finally, the center of gravity of the vascular area is used as the left and right The origin of the carotid artery (points A and B in Figure 3A) and the origin of the left and right vertebral arteries (points C and D in Figure 3A).

9) If the screened area is less than 4, or the screened area is on the side of the center of the human tissue (x direction), the current candidate layer is used as the benchmark, continue to scan upward, and return to S1.1.1 to continue to search for the carotid artery and Origin of the vertebral artery.

Since the vertebral artery originates from the subclavian artery, and the carotid artery originates from the aortic arch and subclavian artery, the origins of the vertebral and carotid arteries are located in the lower half of the eligible CTA image data, so the location of the origin of the carotid and vertebral arteries until CTA The bottom half of the data is scanned layer by layer.

S1.2 Automatically define the endpoints of vertebral and carotid vessels

S1.2.1 Automatically define the end point of the vertebral artery vessel (ie the basilar artery point)

S1.2.1.1 Locate candidate slices of basilar artery point

Scan from top to bottom layer by layer (from the top of the skull to the aortic arch), extract the real brain tissue with the gray value of 120HU as the threshold, and calculate the real brain tissue and the body tissue bounding box after binarization in S1.1.1; When the size of the bounding box of the brain tissue is 1/3 of the bounding box of the body tissue, the scanning is stopped, and this layer is taken as the candidate layer of the basilar artery.

S1.2.1.2 Locate the basilar artery point

In the candidate slice, the area with the gray value of 150-750HU is marked and extracted to obtain the candidate area of the basilar artery, and the following rules are used to screen the candidate area:

1) Calculate the center of gravity of the marked tissue, and remove the unmarked area of the center of gravity;

2) using the blood vessel enhancement function based on the Hessian matrix to perform enhancement processing on the candidate region filtered in step 1);

3) Take 0.6 as the threshold value of blood vessel enhancement, and remove the candidate area corresponding to less than the threshold value;

4) Take the center of gravity of the filtered candidate region in step 3) as the seed point and 100HU as the step size, and carry out two-dimensional region growth;

5) According to the size characteristics of the normal basilar artery, remove the blood vessel candidate area with an area less than 1mm ² and greater than 35mm ² ;

6) Calculate the bounding box of each candidate area, and remove the area with the aspect ratio of the bounding box greater than 4;

7) extracting the boundary of the blood vessel candidate area, calculating the maximum and minimum distances from the center of gravity point to the boundary point, and removing the blood vessel candidate area where the ratio of the maximum distance to the minimum distance is greater than 8;

8) Using the center of gravity of the blood vessel candidate area as the seed point and 100HU as the step size, perform three-dimensional region growth on the upper and lower layers of images, extract the blood vessel area adjacent to the current blood vessel candidate area, and calculate the bounding box of the adjacent blood vessel area; Remove the blood vessel candidate regions with the aspect ratio of the bounding box greater than 4, or the longest side greater than 10mm;

9) Calculate the centering degree, height and area of the blood vessel candidate region after the screening in step 8), wherein the centering degree represents the distance of the center of gravity of the blood vessel candidate region relative to the center of the brain tissue; the height represents the coordinate in the y direction of the center of gravity of the blood vessel candidate region;

10) Sort the three metric parameters calculated in step 9): among them, the centering degree is sorted from small to large, the height and area are sorted from large to small, and the sorted order is used as the metric value of the area. Score: The centering score is expressed as S _center , the height score is expressed as S _y , and the area score is expressed as S _area ;

11) Build a blood vessel similarity score Score=w1*S _center +w2*S _y +w3*S _area , where w1, w1, and w3 represent the weights corresponding to the candidate blood vessel area center score, height score, and area respectively; The centroid of the blood vessel candidate region with the smallest blood vessel similarity score is used as the basilar artery point, such as point E in Figure 3B;

12) If there is no blood vessel candidate area after screening in step 8), take the current layer as the benchmark, continue to scan downward, and return to S1.2.1.1 to continue searching for the base arterial point.

For the CTA image data that meets the requirements, the basilar artery is located in the upper half of the data, so the location of the basilar artery point is scanned layer by layer until the upper half of the CTA data.

S1.2.2 Locate the end point of the carotid artery, that is, the end point of the internal carotid artery

S1.2.2.1 Locating the candidate slice for the end point of the internal carotid artery

Scan layer by layer from top to bottom (from the top of the skull to the direction of the aortic arch), calculate the skull bounding box, construct the intracranial tissue bounding box with the area where the skull bounding box shrinks inward by 1/2, and calculate that the gray value inside the bounding box is greater than 550HU If all the pixels in the bounding box are larger than 550HU, the current layer is determined to be the cranial top area, no judgment is made at this time, and the scan continues downward; when the gray values of all the pixels in the bounding box are equal to Less than 550HU, it is considered that the scan has entered the brain tissue area; when the brain tissue area appears, the ratio of the number of bone pixels in the bounding box to the number of all pixels in the tissue bounding box is determined layer by layer, if the ratio is greater than At 0.005, the search was stopped and this layer was taken as a candidate layer for the end point of the internal carotid artery.

S1.2.2.2 Locate the carotid end point

1) Calculate the gray mean CT _mean and standard deviation CT _std of the blood vessel area where the carotid artery origin is located in S1.1, take the carotid artery origin as the seed point, and take CT _mean + CT _std and CT _mean - CT _std as the The upper and lower thresholds are used for unidirectional upward three-dimensional region growth until the candidate level of the end point of the internal carotid artery, and the candidate blood vessel region in the candidate level is marked at the same time;

2) In the layer where the end point of the internal carotid artery is located, look for the pixel points whose gray value is higher than 750HU, and use it as the seed point of bone tissue, and extract all the threshold values within the range of 120-3071HU that are related to the For the tissue connected by the seed point, remove the candidate blood vessel area that is adhered to the bone in step 1);

3) Calculate the central symmetry degree Symmetry, the bounding box aspect ratio Ratio and the area area of the blood vessel candidate region obtained after the screening in step 2), wherein the definition of the central symmetry degree Symmetry is the same as the definition in step 5) in S1.1.2;

4) Sort the three measurement parameters calculated in step 3): among them, the central symmetry is sorted from small to large, the aspect ratio of the bounding box is sorted from small to large according to the difference from 1, and the area area is sorted from large to large. Sort by small. Take the sorted order as the score of the metric value of the area: the center symmetry score is expressed as S _symmetry , the horizontal and vertical ratio score of the bounding box is expressed as S _ratio , and the area area score is expressed as S _area ;

5) Define the similarity score of blood vessel section: Score=w1*S _symmetry +w2*S _ratio +w3*S _area . Among them, w1, w2, and w3 represent the corresponding weights of the candidate blood vessel region center symmetry score, the bounding box aspect ratio score, and the region area score, respectively. The smaller the blood vessel similarity score, the more likely the area is to be a blood vessel area;

6) Sort the blood vessel cross-section similarity score in the order from small to large, and take the two blood vessel candidate regions with the highest scores as the corresponding center of gravity and define the end points of the left and right carotid arteries according to the x-coordinate direction, as shown in Figure 3C F point and G point of ;

7) If there are less than 2 candidate blood vessel regions after screening in step 2), take the current layer as the benchmark, continue to scan downward, and return to step 2) to continue searching for the end point of the carotid artery.

For the CTA image data that meets the requirements, the end point of the carotid artery is located in the upper half of the data, so for the positioning of the internal carotid artery point until the upper half of the CTA data is scanned slice by slice. .

Step S2 of the present invention: between the start point and the end point of the blood vessel located in step S1, perform multi-scale gradient normalized blood vessel filtering based on Raycasting on the region conforming to the grayscale characteristics of the blood vessel, and construct a blood vessel filtering weight map.

S2.1 Vascular Pre-Extraction

According to the carotid artery and vertebral artery vascular area located in S1.1, the gray mean CT _mean and standard deviation CT _std of the corresponding area were calculated respectively, and the starting point of the carotid artery and vertebral artery located in S1. _mean +2*CT _std and CT _mean -2*CT _std are used as the upper and lower thresholds to perform unidirectional upward three-dimensional regional growth; the carotid artery grows to the level where the end point of the carotid artery is located, and the vertebral artery grows to the level where the end point of the vertebral artery is located;

S2.2 Raycasting-based Multiscale Gradient Normalized Vascular Filtering

In the CTA image, the cross-section of the blood vessel has the characteristics of a circle-like shape, and the grayscale change has the characteristics of a Gaussian-like distribution (the gray value of the center of the blood vessel is higher, and the gray value gradually decreases with the increase of the radius with the center as the center. ). The multi-scale gradient normalized blood vessel filtering based on Raycasting of the present invention utilizes these two characteristics of blood vessels to calculate the blood vessel filtering value of each pixel point in the pre-extracted blood vessel area in S2.1, and the specific steps are as follows:

1) Since the cross-sectional size of the blood vessel will change in its course, it is necessary to calculate the filtering value of the blood vessel under different radius scales (as shown in Figure 2), R ∈ (R _min , R _max ] (in the following steps, R _min and _Rmax represent minimum and maximum radius scales, no longer distinguish between carotid and vertebral arteries);

2) Taking any blood vessel region point pre-extracted in S2.1 as the center of the circle, calculate the projection radius scale as R, the gradient response VR of the center point, and the minimum value VR _,min of the gradient response under the radius scale from 0 to _R , The maximum value VR _,max under the radius scale from R _min to R _max ;

3) Define the normalized gradient response under the radius scale _R as E _R =(VR -VR _,min )/VR _,max , if the current point is the center point of the blood vessel, E _R is approximately equal to 1; if it is a non-vessel point point, E _R is approximately equal to 0;

4) Repeat steps 2)-3), calculate ER under multiple projection radius scales _{R min} to _{R max} , and use the maximum value of ER under different scales as the final blood vessel filter value at this point;

S2.3 Construct blood vessel filtering weight map

For each blood vessel point extracted in S2.1, the filtering value of a blood vessel can be calculated through S2.2, and the blood vessel filtering weight of the blood vessel point is defined as:

in, and are the reciprocal of the blood vessel filter values calculated in S2.2 for two adjacent points, respectively, and Dist is the physical distance between the two points.

According to the calculated blood vessel filter weights of all blood vessel points, a blood vessel filter value weight map is constructed.

Step S3 of the present invention: In the blood vessel filtering weight map constructed in step S2, the Dijkstra optimal path extraction algorithm is used to complete the extraction of the central paths of the carotid artery and the vertebral artery.

Dijkstra's optimal path extraction algorithm is used to obtain the shortest center path between the starting point and the ending point of the blood vessel. First, define the energy function between the starting point and the ending point of the blood vessel located in step S1:

E=∫(w(s)+ε)ds

Among them, w(s) is the blood vessel filter enhancement weight calculated in S2.3, ε is the regular term, and s is the path between the starting point and the ending point.

Then, the path with the smallest sum of the blood vessel filter weights among all the paths between the starting point and the end point of the blood vessel is selected as the center path of the blood vessel.

Figure 3D is the extracted right carotid artery center path and its corresponding CPR (surface reconstruction) result, Figure 3E is the extracted left carotid artery center path and its corresponding CPR (surface reconstruction) result, and Figure 3F is the extracted The central path of the right vertebral artery and its corresponding CPR (curved surface reconstruction) result diagram, Figure 3G is the extracted left vertebral artery central path and its corresponding CPR (curved surface reconstruction) result diagram.

Compared with the prior art, the present invention has the following advantages:

1. Realize the automatic positioning of blood vessels, without the user's manual participation;

2. The extracted centerline is accurate, and no user adjustment is required later;

3. The method runs fast and meets real-time requirements.

Finally, it should be noted that the above-mentioned embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that : it can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements to some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present invention range.

Claims (9)

1. A method for automatically extracting a central path of a head-neck blood vessel in a CTA image is suitable for CTA image data scanned from an aortic arch to a cranial vertex, and is characterized in that: it comprises the following steps:

s1: automatically positioning the starting point and the end point of the carotid artery and the vertebral artery respectively according to the shape, the gray scale and the position characteristics of the blood vessel in the cross section of the CTA image;

s2: performing Raycasting-based multi-scale gradient normalized blood vessel filtering on the region according with the gray level characteristics of the blood vessel between the start point and the end point of the blood vessel positioned in the step S1 to construct a blood vessel filtering weight value graph;

s3: in the vessel filtering weight map constructed in step S2, the Dijkstra optimal path extraction algorithm is used to complete the extraction of the central paths of the carotid artery and vertebral artery vessels.

2. The method of claim 1 for automatically extracting the central path of head and neck blood vessels in CTA image, which comprises: the step S1 is a method for automatically locating the starting points of the carotid artery and the vertebral artery:

s1.1.1 locating candidate levels of carotid and vertebral artery origins

Carrying out binarization preprocessing on CTA image data, and distinguishing human body tissues and background areas in the image; then identifying a hole area with the gray value less than-450 HU in the human tissue area, and scanning layer by layer from bottom to top, when the area of a single hole is more than 60mm²And less than 1200mm²When, consider the current layerThe surface is the upper edge of the lung parenchyma, and the layer is taken as a candidate surface of the starting points of the vertebral artery and the carotid artery at the moment, and the scanning is stopped;

s1.1.2 locating the starting point of vertebral artery and carotid artery

And marking and extracting all regions with the gray values larger than 150HU in the candidate layer to obtain candidate regions of the vertebral artery and the carotid artery blood vessel, screening the candidate regions and determining the starting points of the vertebral artery blood vessel and the carotid artery blood vessel.

3. The method of claim 2 for automatically extracting the central path of the head and neck blood vessels in the CTA image, wherein: the method for screening candidate regions in step S1.1.2 is as follows:

1) calculating the gravity center of the marked tissue, and removing the area of the gravity center which is not marked;

2) on the basis of the step 1), calculating the adjacent edge ratio of the minimum circumscribed rectangle of the remaining candidate region, and removing the region with the ratio being more than 3 or less than 1/3;

3) according to the size characteristics of the normal vertebral artery and the carotid artery, on the basis of the steps 1) and 2), further removing the candidate region with the area larger than 60mm²And less than 3mm²The area of (a);

4) calculating the gray level mean value and the variance of the remaining candidate regions one by one, taking the gravity center of the current candidate region as a seed point, taking the 2-time variance as a growth step length, and carrying out three-dimensional region growth in the upper and lower 3-layer image ranges by taking the layer of the current candidate region as a reference; judging the region grown on each layer again according to the adjacent edge ratio in the step 2), and removing the candidate regions which do not meet the conditions;

5) the cross section of the blood vessel presents a similar circular characteristic in a CTA image, and the distribution in the human body is approximately centrosymmetric, so that the circularity, central Symmetry, bounding box aspect Ratio and Area are respectively calculated for the region screened in the step 4), wherein the central Symmetry represents whether a candidate region exists in the range of 5mm above, below, left and right of the opposite side by taking the center (x direction) of the human tissue after binarization as a reference; if the distance can be found, calculating the distance between the gravity centers of all the candidate blood vessel regions in the search range and the gravity center of the current region in the y direction, and taking the corresponding minimum distance as the central symmetry degree of the candidate blood vessel region; if not, setting the central symmetry of the current blood vessel candidate region as the side length of the whole image;

6) sorting the 4 metric parameters calculated in step 5): the circularity and the area are sorted from large to small, the horizontal-longitudinal ratio is sorted from small to large according to the difference value between the circularity and the area and 1, the symmetry is sorted from small to large, and the sorted order is used as the fraction of the area metric value: the circularity fraction is denoted S_circlarityAnd the symmetry fraction is represented as S_symmetryThe aspect ratio fraction is represented as S_ratioThe area fraction is represented as S_area；

7) Defining a vessel cross-section similarity score: score w 1S_circlarity+w2*S_symmetry+w3*S_ratio+w4*S_areaW1, w2, w3 and w4 respectively represent weights corresponding to the circularity fraction, the symmetry fraction, the aspect ratio fraction and the area fraction of the candidate blood vessel region; the smaller the vessel similarity score is, the more likely the region is a vessel region;

8) sorting the blood vessel similarity Score according to a sequence from small to large to obtain a candidate region ranked in the first four; firstly, defining two regions with larger areas as the initial regions of the left and right carotid arteries, and defining the other two regions as the initial regions of the left and right vertebral arteries; then distinguishing left and right carotid arteries and left and right vertebral arteries according to the obtained x coordinates of the centers of gravity of the carotid artery and vertebral artery initial regions; finally, the gravity center of the blood vessel region is used as the starting points of the left and right carotid arteries and the left and right vertebral arteries;

9) if the screened areas are smaller than 4, or the screened areas are positioned at one side of the center (x direction) of the human tissue, the scanning upwards is continued by taking the current candidate layer as the reference, and simultaneously the starting points of the carotid artery and the vertebral artery are continuously searched by returning to S1.1.1.

4. The method of claim 3 for automatically extracting the central path of the head and neck blood vessels in the CTA image, wherein: the step S1 is a method for automatically positioning the vertebral artery terminal:

the terminal, basilar artery, point of the vertebral artery;

s1.2.1.1 locating basilar artery point candidate slices

Scanning layer by layer (in the direction from the cranial vertex to the aortic arch) from top to bottom, extracting real brain tissue by taking a 120HU gray value as a threshold value, and simultaneously calculating the real brain tissue and the body tissue bounding box after binarization processing in S1.1.1; when the bounding box area of the brain tissue is 1/3 of the volume tissue bounding box, stopping scanning and taking the layer as a candidate layer of the basilar artery;

s1.2.1.2 locating the basilar artery point, i.e. the end point of the vertebral artery

And (4) performing label extraction on the region with the gray value of 150-750HU in the candidate layer to obtain a candidate region of the basilar artery, and screening the candidate region to determine the basilar artery point.

5. The method of claim 4 for automatically extracting the central path of the head and neck blood vessels in the CTA image, wherein: the method for screening the candidate regions to determine the basilar artery points in step S1.2.1.2 is as follows:

1) calculating the gravity center of the marked tissue, and removing the area of the gravity center which is not marked;

2) adopting a blood vessel enhancement function based on a Hessian matrix to perform enhancement processing on the candidate region filtered in the step 1);

3) removing the candidate area corresponding to the threshold value by taking 0.6 as the threshold value of the blood vessel enhancement;

4) taking the gravity center of the candidate region filtered in the step 3) as a seed point and 100HU as a step length, and performing two-dimensional region growth;

5) according to the size characteristics of normal basilar artery, the removal area is less than 1mm²Greater than 35mm²The blood vessel candidate region of (a);

6) calculating a bounding box of each candidate region, and removing regions with the bounding box aspect ratio larger than 4;

7) extracting the boundary of the blood vessel candidate region, calculating the maximum and minimum distances from the gravity point to the boundary point, and removing the blood vessel candidate region of which the ratio of the maximum distance to the minimum distance is more than 8;

8) taking the gravity center of the blood vessel candidate region as a seed point and 100HU as a step length, carrying out three-dimensional region growth on the upper layer image and the lower layer image, extracting a blood vessel region adjacent to the current blood vessel candidate region, and calculating a bounding box of the adjacent blood vessel region; removing the blood vessel candidate area with the aspect ratio of the bounding box being more than 4 or the longest side being more than 10 mm;

9) calculating the centering degree, the height and the area of the blood vessel candidate region screened in the step 8), wherein the centering degree represents the distance between the gravity center of the blood vessel candidate region and the center of the brain tissue; the height represents the coordinate of the gravity center y direction of the blood vessel candidate region;

10) sorting the 3 metric parameters calculated in step 9): the centering degree is sorted from small to large, the height and the area are sorted from large to small, and the sorted order is used as the fraction of the metric value of the region: the median score is denoted S_centerHeight fraction Sy and area fraction S_area；

11) Construction of the vascular similarity Score w 1S_center+w2*S_y+w3*S_areaW1, w1 and w3 respectively represent the median score, height score and weight corresponding to the area of the candidate blood vessel region; taking the gravity center of the blood vessel candidate region with the minimum blood vessel similarity score as a basilar artery point;

12) if no blood vessel candidate area exists after the screening in the step 8), the downward scanning is continued by taking the current layer as a reference, and meanwhile, the method returns to S1.2.1.1 to continue searching for the base artery point.

6. The method of claim 5 for automatically extracting the central path of the head and neck blood vessels in the CTA image, wherein: the step S1 is a method for automatically locating the carotid artery end point:

the end point of the carotid artery is the end point of the internal carotid artery,

s1.2.2.1 locating carotid artery endpoint candidate level

Scanning layer by layer from top to bottom (from the vertex to the aortic arch direction), calculating a skull bounding box, constructing an intracranial tissue bounding box by using an area of the skull bounding box which is shrunk inwards 1/2, and counting the number of pixel points of which the internal gray value is greater than 550 HU; if all pixel points in the range of the bounding box are more than 550HU, judging that the current layer is a cranial vertex region, not judging at the moment, and continuing to scan downwards; when the gray values of all pixels in the bounding box are less than 550HU, the scanning is considered to enter the brain tissue area; after a brain tissue area appears, judging the proportion of the number of bone pixel points in the bounding box to the number of all pixel points in the tissue bounding box in which the bone pixel points are located layer by layer, if the proportion is more than 0.005, stopping searching, and using the layer as a candidate layer of an internal carotid artery endpoint;

s1.2.2.2 locating carotid artery endpoint

1) Calculating the mean value CT of the gray level of the blood vessel region where the carotid artery starting point positioned in S1.1 is located_meanSum standard deviation CT_stdUsing carotid artery starting point as seed point and CT_mean+CT_std、CT_mean-CT_stdPerforming unidirectional upward three-dimensional region growth as an upper threshold and a lower threshold until a candidate layer of the internal carotid artery endpoint is reached, and marking a candidate blood vessel region in the candidate layer;

2) in the layer where the internal carotid artery endpoint is located, searching pixel points with the gray value higher than 750HU, taking the pixel points as bone tissue seed points, extracting all tissues communicated with the seed points within the range of 120-3071HU through a two-dimensional region growing mode, and removing the candidate blood vessel region adhered to the bone in the step 1);

3) calculating the central Symmetry, the bounding box aspect Ratio and the Area of the blood vessel candidate region obtained after screening in the step 2), wherein the Symmetry is the same as the definition of the step 5) in the step S1.1.2;

4) sorting the 3 metric parameters calculated in step 3): the central symmetry degrees are sorted from small to large, the horizontal-vertical ratio of the bounding boxes is sorted from small to large according to the difference value of the horizontal-vertical ratio of the bounding boxes and 1, the area of the region is sorted from large to small, and the sorted order is used as the fraction of the measurement value of the region: the central symmetry fraction is denoted S_symmetryThe enclosure is horizontal and verticalThe ratio score is expressed as S_ratioAnd the area fraction of the region is represented as S_area；

5) Defining a vessel cross-section similarity score: score w 1S_symmetry+w2*S_ratio+w3*S_area(ii) a W1, w2 and w3 respectively represent weights corresponding to the central symmetry fraction, bounding box aspect ratio fraction and region area fraction of the candidate blood vessel region; the smaller the vessel similarity score is, the more likely the region is a vessel region;

6) sorting the vessel section similarity scores Score in a descending order, and defining two vessel candidate regions with the highest scores as corresponding gravity centers as end points of the left and right carotid arteries according to the x coordinate direction;

7) if the candidate blood vessel regions screened in the step 2) are less than 2, continuing to scan downwards by taking the current layer as a reference, and simultaneously returning to the step 2) to continue searching for the stiff artery end point.

7. The method of claim 6 for automatically extracting the central path of the head and neck blood vessels in the CTA image, wherein:

in step S2, a multi-scale gradient normalization blood vessel filtering based on Raycasting is performed on the region conforming to the gray level feature of the blood vessel between the starting point and the ending point of the blood vessel, and the method for constructing the blood vessel filtering weight map is as follows:

s2.1 vascular Pre-extraction

Respectively calculating the mean value CT of the gray levels of the corresponding regions according to the carotid artery and vertebral artery blood vessel regions positioned in S1.1_meanSum standard deviation CT_stdUsing carotid and vertebral artery starting points located in S1.1 as seed points, CT_mean+2*CT_std、CT_mean-2*CT_stdPerforming unidirectional upward three-dimensional region growth as an upper threshold and a lower threshold; the carotid artery grows to the level of the carotid artery terminal point, and the vertebral artery grows to the level of the vertebral artery terminal point;

s2.2, based on Raycasting multi-scale gradient normalization blood vessel filtering, calculating a blood vessel filtering value of each pixel point in the pre-extracted blood vessel region in S2.1;

s2.3 construction of vascular filtering weight map

For each blood vessel point extracted in S2.1, a blood vessel filtering value can be calculated through S2.2, and a blood vessel filtering weight value of the blood vessel point is defined as:

wherein,andrespectively calculating the reciprocal of the blood vessel filtering value calculated in S2.2 for two adjacent points, and Dist is the physical distance between the two points;

and constructing a blood vessel filtering value weight map according to the calculated blood vessel filtering weights of all the blood vessel points.

8. The method of claim 7 for automatically extracting the central path of the head and neck blood vessels in the CTA image, wherein: s2.2, based on Raycasting multi-scale gradient normalization blood vessel filtering, calculating a blood vessel filtering value of each pixel point in a pre-extracted blood vessel region in S2.1, and specifically comprising the following steps:

1) since the cross-sectional dimension of a blood vessel varies in its course, it is necessary to calculate the filtered value of the blood vessel at different radius scales, R e (R ∈)_min,R_max](in the subsequent step with R_minAnd R_maxRepresent the minimum and maximum radius dimensions, no longer distinguishing between carotid and vertebral arteries);

2) taking any one blood vessel region point pre-extracted in S2.1 as a circle center, respectively calculating the projection radius scale as R and the gradient response V of the circle center point_RMinimum V of gradient response on the 0 to R radius scale_R,min，R_minTo R_maxMaximum value V at radius scale_R,max；

3) Defining the normalized gradient response at the radius scale R as E_R＝(V_R-V_R,min)/V_R,maxIf the current point is the center point of the blood vessel, E_RAbout equal to 1; if it is a non-vascular point, E_RAbout 0;

4) repeating steps 2) -3), calculating a plurality of projection radius scales R_minTo R_maxE of_RAnd will be at different scales E_RThe maximum value of (a) is used as the final blood vessel filtering value at the point.

9. The method of claim 8 for automatically extracting the central path of the head and neck blood vessels in the CTA image, wherein: the step S3: in the vessel filtering weight map constructed in step S2, the method for extracting the central paths of the carotid artery and the vertebral artery by using Dijkstra optimal path extraction algorithm is as follows:

s3.1: defining an energy function between the start and end points of the blood vessel located in step S1:

E＝∫(w(s)+ε)ds

wherein w (S) is the blood vessel filtering enhancement weight calculated in S2.3, epsilon is a regular term, and S is a path between a starting point and an end point;

and S3.2, selecting the path with the minimum blood vessel filtering weight sum in all paths between the starting point and the end point of the blood vessel as the blood vessel central path.