Background
Common morbidity modes of cardiovascular diseases include hypertension, coronary heart disease, myocardial infarction, cerebral hemorrhage and the like. Among them, most of the cardiovascular and cerebrovascular diseases are based on the distortion of the blood vessel wall. Atherosclerosis and high-pressure arteriosclerosis cause thickening and hardening of blood vessel walls, loss of elasticity, reduction of lumen, complete occlusion, pathological changes and cardiovascular and cerebrovascular diseases. Cardiovascular and cerebrovascular diseases have the characteristics of high morbidity, high mortality, high disability rate and the like, and are main chronic non-infectious diseases affecting the life health of human beings. Therefore, it is necessary to prevent the occurrence or recurrence of cardiovascular and cerebrovascular diseases by detecting and diagnosing risk factors in blood vessels in advance, thereby achieving the purpose of reducing the mortality.
Diagnosis of morphology of blood vessels as one of the methods of diagnosis of blood vessels, it is necessary to detect and image blood vessels by means of devices and methods such as CT (computed tomography), MRA (magnetic resonance angiography), angiography, or angioscope before diagnosis. When the device is used for detecting and imaging the blood vessel, the obtained initial image not only contains the blood vessel, but also contains other tissues around the blood vessel and interference signals. Since the blood vessel is surrounded by other surrounding tissues and the surrounding tissue images may affect the observation of the blood vessel tissue by the doctor, the initial image needs to be processed and segmented by a large number of image processing means to separate the blood vessel signals and reconstruct the blood vessel signals into the blood vessel image. The target region of the blood vessel image segmentation is a blood vessel signal, and other parts are all regarded as background regions. The effective segmentation method aims to distinguish blood vessels from surrounding tissues, and blood vessels can be observed more clearly after interference is removed, so that diagnosis can be performed more accurately. And after the blood vessel is segmented, the parameters such as the sectional area of the blood vessel and the like can be more conveniently and quantitatively measured. Therefore, accurate segmentation of blood vessels is of great significance for three-dimensional imaging of blood vessels.
The Chinese invention patent with the application number of 201710775038.X provides a blood vessel image segmentation method and a nuclear magnetic resonance imaging system based on centerline extraction. The method comprises the steps of preprocessing brain blood vessel data by adopting vesselness filtering based on a Hessian matrix, extracting a blood vessel central line by adopting a topology thinning method, carrying out edge expansion on an original image, extracting the characteristics of a training sample and a test sample by taking the central line as a positive sample and taking a non-blood vessel point as a negative sample, and training an SVM model to obtain a blood vessel segmentation result. The method does not need to manually calibrate the target and the background, and realizes full-automatic blood vessel segmentation.
However, since the cardiovascular and cerebrovascular vessels have extremely complicated structures, the sizes and curvatures thereof are variable, there are multi-level branches in a small range, the branches at the ends are very fine, and the appearance and the geometric features of the diseased vessels are affected by the conditions of vessel expansion, calcification, aneurysm, stenosis, etc., resulting in increased difficulty of vessel segmentation based on centerline extraction and low segmentation accuracy.
Disclosure of Invention
In view of the above-mentioned drawbacks of the prior art, it is an object of the present invention to provide an imaging system and method for cardiovascular and cerebrovascular detection. The imaging system pre-divides the two-dimensional image of the cross section of the blood vessel through an image pre-dividing module, preliminarily extracts the blood vessel signal of each layer, then performs secondary division on the blood vessel signal of each layer through an image secondary dividing module, filters interference signals to obtain the blood vessel image of each layer, and finally performs three-dimensional conversion to obtain the three-dimensional image of the blood vessel. The invention combines image pre-segmentation and secondary segmentation, gradually filters interference signals except blood vessels, and finally obtains a high-precision blood vessel three-dimensional image.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
an imaging system for cardiovascular and cerebrovascular detection comprises an image acquisition module, an image pre-segmentation module, an image secondary segmentation module, an image conversion module and a display module which are sequentially connected;
the image acquisition module is used for receiving or acquiring two-dimensional images of the cross section of the blood vessel of different layers;
the image pre-segmentation module is used for pre-segmenting the two-dimensional image of the cross section of the blood vessel and preliminarily extracting a blood vessel signal of each layer;
the image secondary segmentation module is used for carrying out secondary segmentation on the blood vessel signals of each layer, filtering interference signals and obtaining blood vessel images of each layer;
the image conversion module is used for converting the blood vessel image of each layer into a blood vessel three-dimensional image;
the display module is used for displaying the blood vessel three-dimensional image.
Furthermore, the image pre-segmentation module pre-segments the two-dimensional image of the cross section of the blood vessel through a finite mixed normal distribution image pre-segmentation model.
Further, the finite mixture normal distribution image pre-segmentation model solves model parameters through an expected maximization algorithm.
Further, the secondary segmentation adopts an image segmentation model to perform secondary segmentation on the blood vessel signal of each slice, where the image segmentation model is expressed as follows:
in the formula, I represents the set of all superpixels, omega represents the set of all neighborhood superpixel pairs in the same superpixel layer, C represents the set of corresponding superpixel pairs between different superpixel layers, and Di(fi) Is a super pixel i and its label fiλ and φ are balance parameters, Bmn(fm,fn) Is a penalty function, S, between the neighbourhood superpixels m and n in the same superpixel layerpq(fp,fq) A function is measured for similarity of corresponding superpixel pairs between different superpixel layers.
Further, the construction method of the image segmentation model is as follows:
(1) generating three layers of super pixels with different scales of the blood vessel signal by using a super pixel segmentation algorithm;
(2) extracting the characteristics of the superpixels in the step (1) and acquiring the structural information of the superpixels;
(3) introducing the structural information of the superpixel in the step (2) into an image segmentation model based on graph segmentation to obtain the image segmentation model; the graph cut-based image segmentation model is shown as follows:
where I represents the set of all superpixels, ω represents the set of all neighborhood superpixel pairs within the same superpixel layer, Di(fi) Is a super pixel i and its label fiλ is a balance parameter, Bmn(fm,fn) Is a penalty function between neighboring superpixels m and n within the same superpixel layer.
Further, in the step (1), the number of the super pixels with different three-layer dimensions is respectively 300-.
Further, in step (2), the structural information of the super-pixel includes a mean and a covariance of the pixel characteristics.
Further, the image conversion module stacks the blood vessel images of each slice by a maximum intensity projection method to obtain the blood vessel three-dimensional image.
Further, the image acquisition module acquires two-dimensional images of cross sections of the blood vessels of different layers through CT equipment or magnetic resonance equipment.
An imaging method of the imaging system for detecting the cardiovascular and cerebrovascular vessels comprises the following steps:
s1, the image acquisition module acquires two-dimensional images of cross sections of blood vessels on different layers;
s2, the image pre-segmentation module pre-segments the two-dimensional image of the cross section of the blood vessel of each layer in the step S1 respectively, and preliminarily extracts a blood vessel signal of each layer to obtain an initial blood vessel image of each layer;
s3, performing secondary segmentation on the initial blood vessel image of each layer in the step S2 by the image secondary segmentation module to obtain a blood vessel image of each layer;
and S4, the image conversion module stacks the blood vessel images of each layer in the step S3 in sequence to obtain a blood vessel three-dimensional image, and the blood vessel three-dimensional image is displayed through the display module.
Advantageous effects
Compared with the prior art, the imaging system and the method for detecting the cardiovascular and cerebrovascular vessels have the following beneficial effects:
(1) the imaging system for detecting the cardiovascular and cerebrovascular vessels comprises an image acquisition module, an image pre-segmentation module, an image secondary segmentation module, an image conversion module and a display module which are sequentially connected. The image pre-segmentation module pre-segments the two-dimensional image of the cross section of the blood vessel by adopting a finite mixed normal distribution image pre-segmentation model, and preliminarily extracts a blood vessel signal of each layer; secondly, performing secondary segmentation on the blood vessel signal of each layer by an image segmentation model by an image secondary segmentation module, and filtering interference signals to obtain a blood vessel image of each layer; and finally, the image conversion module converts the blood vessel image of each layer into a blood vessel three-dimensional image by adopting a maximum density projection method, so that the three-dimensional imaging of the cardiovascular and cerebrovascular is realized. According to the program system, the invention combines image pre-segmentation and secondary segmentation to gradually filter out interference signals except blood vessels, wherein the image pre-segmentation model adopts finite mixed normal distribution to obtain blood vessel data through preliminary segmentation, the image segmentation model extracts three superpixels with different scales of the blood vessel data through preliminary segmentation, interference signals (such as adipose tissues) similar to the gray scale of the blood vessel are further filtered, and finally a high-precision blood vessel three-dimensional image is obtained.
(2) The image pre-segmentation module adopts a finite mixed normal distribution image pre-segmentation model, and the construction of the model is realized by assuming that a medium brightness area in a gray distribution histogram of a two-dimensional image of a cross section of a blood vessel is formed by two normal distributions and a total distribution curve is formed by four components. The gray distribution histogram of the two-dimensional image of the cross section of the blood vessel can be divided into three main component brightness levels, the lowest level brightness is a noise signal, cerebrospinal fluid and bone tissues; the medium luminance signal is a signal of brain tissue, including signals of gray and white matter; the highest brightness signal component is mainly the signal of arterial blood flow and adipose tissue around the brain. Verification shows that the assumption accords with the true condition of blood vessel image distribution, and blood vessel signal components can be well separated from mixed components, so that the purpose of preliminarily segmenting the blood vessel image is achieved.
(3) The image segmentation module firstly extracts the superpixels of three different scales of the blood vessel signals obtained by primary segmentation, then extracts the mean value and covariance of the pixel characteristics of the superpixels as the structural information of the superpixels, and finally introduces the superpixel structural information into an image segmentation model based on image segmentation to obtain the image segmentation model. According to the image segmentation model, the superpixel information of three different scales is introduced into the model, the small-scale superpixel can store the local information of the image, and local adjacent constraint can be provided to overcome over-segmentation; the large-scale superpixel can store the macrostructure information of the image, and can provide the regional connectivity constraint of the long range of the image to overcome under-segmentation; the super-pixel of the medium-dimensional scale can transfer image global information. By fusing the three scales of super-pixel information, the quality of the segmentation of the blood vessel image is further improved, so that a high-precision blood vessel three-dimensional image is obtained.
Detailed Description
The technical solutions of the embodiments of the present invention will be described clearly and completely below, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all embodiments; all other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without any inventive step, are within the scope of the present invention.
Referring to fig. 1, the imaging system for detecting cardiovascular and cerebrovascular diseases provided by the present invention includes an image acquisition module, an image pre-segmentation module, an image secondary segmentation module, an image conversion module and a display module, which are connected in sequence.
The image acquisition module is used for receiving or acquiring the two-dimensional images of the cross section of the blood vessel of different layers.
Preferably, the image acquisition module acquires two-dimensional images of cross sections of the blood vessels of different layers through a CT device or a magnetic resonance device. Taking magnetic resonance blood vessel imaging equipment as an example, magnetic resonance imaging is a common diagnostic means, and has the advantages of good tissue contrast, no ionizing radiation and the like. In clinic, a patient is placed in the center of a magnet, hydrogen atoms (common condition) in the body of the patient are subjected to space encoding by controlling gradient pulses and radio frequency pulses, signals are collected to obtain k-space data, and then the k-space data are subjected to two-dimensional Fourier transform to obtain a magnetic resonance image. Magnetic resonance angiography mainly utilizes a saturation effect, an inflow enhancement effect and a flow dephasing effect. The magnetic resonance blood vessel acquisition principle is as follows: and transmitting a pulse signal to a certain layer for multiple times to enable the tissue of the layer to reach a saturation state, and acquiring an image of the layer after a period of time TR. The static tissue is in a saturated state, the signal is low, the blood in the blood vessel is in a flowing state, the blood at the layer surface position flows in from other positions, and the high signal is presented after the saturation. Therefore, the blood and the peripheral tissue signals are contrasted to reflect that the blood vessels are highlighted on the image and the peripheral static tissues are darker, thereby realizing the acquisition of the blood vessel image. It can be seen that the initial image acquired by magnetic resonance angiography is a vessel cross-section two-dimensional image.
The image pre-segmentation module is used for pre-segmenting the two-dimensional image of the cross section of the blood vessel and preliminarily extracting the blood vessel signal of each layer.
Furthermore, the image pre-segmentation module pre-segments the two-dimensional image of the cross section of the blood vessel through a finite mixed normal distribution image pre-segmentation model. The finite mixed normal distribution image pre-segmentation model is shown as the formula (1):
in the formula, a
tIs f
t(x) Weight of (f)
t(x) Is the t-th distribution in the mixed normal distribution
T is 1,2,3,4, f
4For mixed distribution of vascular components, f
1,f
2,f
3For mixed distribution of background components,. phi
kIs a parameter vector of the mixed distribution density function, as shown in formula (2):
further, 12 unknown parameters in the finite mixture normal distribution image pre-segmentation model
Solved by the expectation maximization algorithm.
Then according to the maximum posterior classification method, if formula (3) is satisfied:
then voxel x is considerediBelonging to the blood vessel.
The limited mixed normal distribution model is that in the gray distribution histogram of the two-dimensional image of the cross section of the assumed blood vessel, the medium brightness area is formed by two normal distributions, and the total distribution curve is formed by four components. The grey level distribution histogram of the two-dimensional image of the cross section of the blood vessel can be divided into three main component brightness levels. The lowest level of brightness is the noise signal, cerebrospinal fluid and bone tissue; the medium luminance signal is a signal of brain tissue, including signals of gray and white matter; the highest brightness signal component is mainly the signal of arterial blood flow and adipose tissue around the brain, and fat is high signal like blood flow due to the short T1 relaxation time of fat. Tests show that the hypothesis is relatively consistent with the real situation, and the blood vessel signal component can be well separated from the mixed component, so that the purpose of preliminarily segmenting the blood vessel image is achieved.
And the image secondary segmentation module is used for carrying out secondary segmentation on the blood vessel signals of each layer, filtering interference signals and obtaining the blood vessel image of each layer.
Further, the secondary segmentation adopts an image segmentation model to perform secondary segmentation on the blood vessel signal of each slice, and the image segmentation model is as shown in formula (5):
in the formula, I represents the set of all superpixels, omega represents the set of all neighborhood superpixel pairs in the same superpixel layer, C represents the set of corresponding superpixel pairs between different superpixel layers, and Di(fi) Is a super pixel i andits label fiλ and φ are balance parameters, Bmn(fm,fn) Is a penalty function, S, between the neighbourhood superpixels m and n in the same superpixel layerpq(fp,fq) A function is measured for similarity of corresponding superpixel pairs between different superpixel layers.
Further, the construction method of the image segmentation model is as follows:
(1) generating three layers of super pixels with different scales of the blood vessel signal by using a super pixel segmentation algorithm;
the pixel numbers of the super pixels with different three-layer scales are respectively 300-500, 800-1500 and 6000-10000; wherein the small-scale superpixel stores local information of the blood vessel image, and can provide local adjacent constraint to overcome over-segmentation; the large-scale superpixel stores the macrostructure information of the image, and can provide the regional connectivity constraint of the long range of the blood vessel image so as to overcome under-segmentation; the super-pixel of the medium-dimensional scale can transfer the global information of the blood vessel image.
(2) Extracting the characteristics of the superpixels in the step (1) and acquiring the structural information of the superpixels;
the structural information of the super-pixel comprises a mean value and a covariance of pixel characteristics; e.g. a super-pixel i can be characterized as Gi={μi,∑i},μiIs the mean value, Σ, of the super pixel iiIs the covariance of the superpixel i, the feature G of the superpixel mm={μm,∑mAnd the feature G of the super-pixel nn={μn,∑nThe definition of similarity between them is shown in formula (6):
where tr (-) is the trace operation of the matrix and d is the dimension of the superpixel feature.
(3) Introducing the structural information of the superpixel in the step (2) into an image segmentation model based on graph segmentation to obtain an image segmentation model shown in a formula (5); the image segmentation model based on graph cut is shown as formula (7):
where I represents the set of all superpixels, ω represents the set of all neighborhood superpixel pairs within the same superpixel layer, Di(fi) Is a super pixel i and its label fiλ is a balance parameter, Bmn(fm,fn) Is a penalty function between neighboring superpixels m and n within the same superpixel layer.
The invention adopts the component form expectation maximization Gaussian mixture algorithm to perform clustering solution on the image segmentation model, and clusters the foreground seeds and the background seeds to obtain corresponding Gaussian sets
And
where N and M are the number of classes of foreground and background, respectively. The similarity between the superpixel i and the foreground and background models is shown as equation (8) and equation (9):
Bmn(fm,fn) For the penalty function between neighboring superpixels m and n within the same superpixel layer, the following formula (10) is defined:
Bmn(fm,fn)=exp(-ξmdis2(Gm,Gn))·(fm≠fn) (10)
in the formula (f)m≠fn) 1 if fm≠fnOtherwise, it is 0. Adjusting parameter ximThe method is set based on the super-pixel layers to which m and n belong, namely each super-pixel layer has respective adjusting parameter values, and the specific definition is as shown in a formula (11):
in the formula, | omegamAnd | represents the number of elements of the super-pixel layer to which m belongs. Spq(fp,fq) Defining a similarity measure function for corresponding superpixel pairs between different superpixel layers as shown in equation (12):
Spq(fp,fq)=exp(-ξmdis2(Gp,Gq))·(fp≠fq) (12)
in the formula (f)p≠fq) 1 if fp≠fqOtherwise, it is 0. Regulating parameter value xi of super pixel layer to which p belongspAnd the adjusting parameter value xi of the super pixel layer to which q belongsqAre not equal, and therefore the adjustment parameter xi between different super pixel layerspqThe definition is shown in formula (13):
and the image conversion module is used for converting the blood vessel image of each layer into a blood vessel three-dimensional image.
Further, the image conversion module stacks the blood vessel images of each slice by a maximum intensity projection method to obtain the blood vessel three-dimensional image. The idea of the method is to emit rays at a projection angle for each pixel in the projection plane, pass through the volume data, compare the data values on each ray, and select the largest value as the display value of the response pixel. The maximum density projection method has the main advantages of small calculated amount and high calculating speed.
The display module is used for displaying the blood vessel three-dimensional image.
Referring to fig. 2, the imaging method of the imaging system for detecting cardiovascular and cerebrovascular vessels described above includes the following steps:
s1, the image acquisition module acquires two-dimensional images of cross sections of blood vessels of different layers through CT equipment or magnetic resonance imaging equipment;
s2, the image pre-segmentation module pre-segments the two-dimensional image of the cross section of the blood vessel of each layer in the step S1 by adopting a finite mixed normal distribution image pre-segmentation model, preliminarily extracts the blood vessel signal of each layer and obtains an initial blood vessel image of each layer;
s3, the image secondary segmentation module carries out secondary segmentation on the initial blood vessel image of each layer in the step S2 by adopting an image segmentation model, and further filters interference signals to obtain a blood vessel image of each layer;
and S4, the image conversion module adopts a maximum density projection method to stack the blood vessel images of each layer in the step S3 in sequence to obtain a blood vessel three-dimensional image, and the blood vessel three-dimensional image is displayed through the display module.
In summary, the imaging system for detecting cardiovascular and cerebrovascular vessels provided by the invention comprises an image acquisition module, an image pre-segmentation module, an image secondary segmentation module, an image conversion module and a display module which are sequentially connected. The image pre-segmentation module pre-segments the two-dimensional image of the cross section of the blood vessel by adopting a finite mixed normal distribution image pre-segmentation model, and preliminarily extracts a blood vessel signal of each layer; secondly, performing secondary segmentation on the blood vessel signal of each layer by an image segmentation model by an image secondary segmentation module, and filtering interference signals to obtain a blood vessel image of each layer; and finally, the image conversion module converts the blood vessel image of each layer into a blood vessel three-dimensional image by adopting a maximum density projection method, so that the three-dimensional imaging of the cardiovascular and cerebrovascular is realized. According to the program system, the invention combines image pre-segmentation and secondary segmentation to gradually filter out interference signals except blood vessels, wherein the image pre-segmentation model adopts finite mixed normal distribution to obtain blood vessel data through preliminary segmentation, the image segmentation model extracts three superpixels with different scales of the blood vessel data through preliminary segmentation, interference signals (such as adipose tissues) similar to the gray scale of the blood vessel are further filtered, and finally a high-precision blood vessel three-dimensional image is obtained.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.