CN113706494A - Full-automatic medical image segmentation method and device based on shape prior - Google Patents

Full-automatic medical image segmentation method and device based on shape prior Download PDF

Info

Publication number
CN113706494A
CN113706494A CN202110966535.4A CN202110966535A CN113706494A CN 113706494 A CN113706494 A CN 113706494A CN 202110966535 A CN202110966535 A CN 202110966535A CN 113706494 A CN113706494 A CN 113706494A
Authority
CN
China
Prior art keywords
image
shape prior
shape
prior
segmentation
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110966535.4A
Other languages
Chinese (zh)
Inventor
贵鹿颖
马骏
杨孝平
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN202110966535.4A priority Critical patent/CN113706494A/en
Publication of CN113706494A publication Critical patent/CN113706494A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/12—Edge-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10101—Optical tomography; Optical coherence tomography [OCT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30101—Blood vessel; Artery; Vein; Vascular

Abstract

The invention discloses a full-automatic medical image segmentation method and device based on shape prior, which are suitable for OCT image vascular wall separation. The method comprises the following steps: selecting an OCT image to be processed, converting original OCT data into a rectangular coordinate system, and extracting a part with the maximum gray value as a reference image; detecting the outer wall of the conduit and generating the outer wall surface of the conduit; taking the outer wall surface of the conduit as initialization and shape prior, and fusing the initialization and the shape prior into a variation frame to obtain a geometric active contour model based on the shape prior; and solving the active contour model, updating the shape prior according to the obtained segmentation curve, and alternately performing the operation of solving the model and updating the shape prior until an iteration stop condition is reached to obtain a segmentation result. The invention can automatically generate the shape prior of the blood vessels with different radiuses, thereby effectively segmenting the inner wall of the blood vessel on the OCT image interfered by factors such as boundary deletion, blood, artifact and the like without manual intervention or optimization of shape parameters.

Description

Full-automatic medical image segmentation method and device based on shape prior
Technical Field
The invention belongs to the technical field of medical image processing, and particularly relates to a full-automatic medical image segmentation method and device based on automatically updated shape prior.
Background
The main cause of coronary heart disease is atherosclerosis lesion which causes coronary artery lumen narrowing, and further causes artery occlusion, which is life threatening. Medically, the degree of stenosis is a key index for evaluating the health of blood vessels, and the detection of the degree of stenosis is mainly realized by calculating the area of the blood vessels. In addition, luminal area information is also critical to the selection and placement of coronary stent size in clinical treatment. Accurate segmentation of lumen/vessel walls on medical images can provide accurate lumen area information.
A number of imaging methods are used to assess the state and extent of stenosis of the coronary arteries. The main conventional imaging methods include angiography, intravascular ultrasound (IVUS), intravascular Optical Coherence Tomography (IVOCT). OCT has become a widely used imaging technique in clinical practice because it can provide high-resolution images and display the internal structure of blood vessels.
The imaging quality of OCT is often closely related to the operator experience and the imaging equipment, and in practical image segmentation, challenging problems still exist. First, there is always a partial absence of the lumen boundary in the OCT image, since the shadow of the guidewire inevitably appears in each image. In addition, partial deletion of the target boundary can be caused by too far distance from the imaging guide wire, occlusion and the like. Secondly, there is also interference information in the image. Some false objects, such as residual blood and guide wires, have similar gray features to the vessel wall, thereby affecting the accuracy of segmentation. Third, the vessels exhibit severe fluctuations in the OCT images due to motion artifacts caused by the high-speed rotational advancement of the guidewire. It is worth noting that the above problems occur almost in every image instance, and because of the different operator levels, with different degrees of image degradation, it is important to deal with these problems for accurately segmenting the lumen.
The OCT imaging equipment images in blood vessels, the obtained original data are images of the blood vessels under polar coordinates, and some segmentation algorithms directly segment lumen boundaries on the data through methods such as classification, active contour lines, clustering and the like. However, the original data in polar coordinates cannot provide geometric information of the blood vessel, and these data are usually converted into cartesian coordinates to segment the image by thresholding, dynamic programming, Dijkstra's algorithm, level set method, interpolation method, etc. However, most of the above algorithms are performed on OCT images with good image quality, and segmentation on low-quality images is rarely involved. In addition, for interference factors commonly existing in the OCT image, such as partially missing boundaries, guidewire shadows, false boundaries caused by residual blood, and the like, some algorithms employ a preprocessing step to process the situations, such as filling up the missing boundaries, and the like, which increases additional processing cost and has poor interference resistance. In recent years, methods based on deep learning are also used to segment the lumen on OCT images and obtain relatively good segmentation results, but these methods usually require a large number of labeled images as training sets to ensure the accuracy of segmentation, and these labeled images are expensive and difficult to obtain.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide a medical image segmentation method, which solves the problems of unsupportability to low-quality images, poor anti-interference capability, high processing cost and the like in the existing OCT image vessel wall segmentation method, and achieves the purpose of accurately segmenting the vessel wall.
The invention also provides medical imaging equipment for implementing the medical image segmentation method.
The technical scheme is as follows: in order to achieve the above purpose, the invention adopts the following technical scheme:
in a first aspect, a method for fully automatic medical image segmentation based on shape prior comprises the following steps:
(1) selecting a three-dimensional OCT image to be processed, converting original OCT data into a rectangular coordinate system, and extracting a part with the maximum gray value as a reference image;
(2) detecting the outer wall of the catheter according to the three-dimensional OCT image, and generating the outer wall surface of the catheter;
(3) initializing the outer wall surface of the conduit as a curve and shape prior, and fusing the curve and the shape prior into a variational frame to obtain a geometric active contour model based on the shape prior;
(4) and solving the active contour model, updating the shape prior according to the obtained segmentation curve, and alternately performing the operation of solving the model and updating the shape prior until an iteration stop condition is reached to obtain a segmentation result.
According to certain embodiments of the first aspect, the step (2) specifically comprises: extracting two vertical sections which are vertical to each other of the three-dimensional OCT image, respectively calculating the gradient of each vertical section in the y-axis direction, and accumulating the gradient image in the x-axis direction; clustering the upper part and the lower part of the accumulated result respectively, and selecting the one with larger value from the two types, wherein the point farthest from the center of the image in the one type is the position of the outer wall of the catheter; obtaining two straight line positions in each longitudinal section image, and then obtaining four straight line positions in the two longitudinal sections; and correspondingly returning the linear coordinates to the three-dimensional image to obtain four straight lines in the three-dimensional image, and calculating the cylindrical surfaces of the four straight lines to obtain a standard cylindrical surface, namely the outer wall surface of the catheter.
According to certain embodiments of the first aspect, the geometric active contour model based on shape priors in step (3) is represented using the following total energy functional:
E＝EGAC+Eprior+Eregional (*)
wherein E isGACEnergy of geometric active contour line model, EpriorBeing the shape prior energy, EregionalIs the regional energy;
the energy calculation formula of the geometric active contour line model is as follows:
Figure BDA0003224128710000038
wherein, alpha and lambda are the weights of two energy terms, phi is a level set function, H is a Heaviside function, and g is a boundary operator; omega is the image area, delta is the Dirac function,
Figure BDA0003224128710000031
is a gradient operator, which is a linear operator,
Figure BDA0003224128710000032
is the gradient of phi;
the shape prior energy calculation formula is as follows:
Figure BDA0003224128710000033
where β is the weight of the shape prior energy,
Figure BDA0003224128710000034
is a level set function of shape priors; r, c are shape prior parameters, r is radius, c is centerline;
the calculation formula of the region energy is as follows:
Eregional＝γ∫Ω(I-f1)2H(Φ)+(I-f2)2(1-H(Φ))dx
where γ is the weight of the region energy, I is the gray level of the image, f1，f2The average values of the image gradations of the regions where H is 1 and H is 0, respectively.
According to certain embodiments of the first aspect, in step (4), solving the active contour model comprises: solving the total energy functional using a fast algorithm that uses a characteristic function u (x) to represent the segmented regions, the characteristic function being:
Figure BDA0003224128710000035
where Γ is the segmentation curve of the image, ΩΓAn inner region representing a curve Γ;
the total energy functional (×) is approximately represented as follows:
Figure BDA0003224128710000036
wherein the content of the first and second substances,
Figure BDA0003224128710000037
is a shape prior represented by a characteristic function;
find out Eτ(u) to a minimum of u, i.e. an optimal segmentation result.
According to certain embodiments of the first aspect, the updating of the shape prior according to the obtained segmentation curve in step (4) comprises: defining the height of all cylinders as a pixel, determining a central line and a radius according to the position of a segmentation curve on each frame in an image, determining cylinders according to the central line and the radius, comparing each newly generated cylinder with a reference image, recording the radius of the cylinder which does not intersect with the reference image, and calculating the average value of the radius of the cylinders to serve as the uniform radius of the shape prior.
In a second aspect, a medical imaging device comprises: memory storing one or more computer programs which, when executed by one or more processors, cause the one or more processors to perform steps comprising the fully automatic shape-prior based medical image segmentation method of the first aspect of the invention.
Has the advantages that: the invention discloses a full-automatic medical image segmentation method and equipment based on shape prior.A to-be-processed OCT three-dimensional image is firstly detected on the outer wall of a catheter, and the outer wall surface of the catheter is generated; taking the outer wall surface of the conduit as initialization and shape prior, and fusing the initialization and the shape prior into a variation frame to obtain a geometric active contour model based on the shape prior; and solving the active contour model to obtain the segmentation of the image. The invention can automatically generate the shape prior of the blood vessels with different radiuses, can effectively segment the inner wall of the blood vessel on the OCT image interfered by factors such as boundary deletion, blood, artifact and the like, and does not need manual intervention or optimization of shape parameters.
Drawings
FIG. 1 is a flow chart of a fully automated medical image segmentation method based on shape priors according to an embodiment of the invention;
FIG. 2 is a schematic view of outer catheter line detection on a longitudinal section according to an embodiment of the present invention;
FIG. 3 is a flow chart of the generation of an initialized standard cylindrical surface on a three-dimensional OCT image according to an embodiment of the invention;
FIG. 4 is a schematic diagram of an initialized standard cylindrical surface generated on a three-dimensional OCT image according to an embodiment of the invention;
FIG. 5 is a comparison of segmentation results for different models on less noisy images, according to an embodiment of the invention;
FIG. 6 is a comparison of segmentation results for different models on a more noisy image, according to an embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
Referring to fig. 1, the method for generating lumen segmentation on adaptive shape prior three-dimensional OCT images provided by the present invention includes the following steps:
and (1) inputting a three-dimensional OCT image to be segmented, and converting original data into a rectangular coordinate system. The quadratic Otsu threshold is calculated, and the part with the highest gray value is taken as the reference image of the example.
And (2) detecting the outer wall of the conduit and generating the outer wall surface of the conduit.
Two vertical sections which are vertical to each other are taken on the three-dimensional OCT image, the gradient of the y-axis direction is calculated for each vertical section, and then the gradient is accumulated along the x-axis direction to obtain a group of one-dimensional data. The group of data is divided into two groups, namely k-means clustering and k-means clustering. The class with the larger data value (the class with the fewer objects in the class) is taken, the positions of the data points in the class are arranged, and the point farthest from the center is used as the target point. FIG. 2 is a schematic view of outer catheter line detection on a longitudinal section according to an embodiment of the present invention; wherein (a) is a longitudinal section of a three-dimensional OCT image; (b) is (a) a gradient plot along the Y-axis direction; (c) is the summation of the gradient map (b) along the x-axis direction.
The y-axis coordinate of the target point is the position of the corresponding detection straight line on the longitudinal section, namely the straight line on which the outer wall of the catheter is located. In this way, four straight lines can be detected on the two longitudinal sections. Placing them back in the three-dimensional OCT image enables a standard cylindrical surface to be determined. This is the initialization of the shape prior and the segmentation curve. FIG. 3 is a flow chart of the generation of an initialized standard cylindrical surface on a three-dimensional OCT image according to an embodiment of the invention, where (a) is the raw data of the OCT image in polar coordinates; (b) is an image converted to rectangular coordinates; (c1) (c2) are two longitudinal sections perpendicular to each other, respectively, and the straight line on which the outer wall of the duct is found on the longitudinal sections; (d) four straight lines are placed back into the three-dimensional image to generate a cylindrical surface, and the generated result is used as the initialization of the active contour line and the initialization of the shape prior as shown in fig. 4.
And (3) initializing the outer wall surface of the conduit as a curve and shape prior, and fusing the curve and the shape prior into a variation frame to obtain a geometric active contour model based on the shape prior.
The geometry activity profile model based on shape priors is represented using the following total energy functional:
E＝EGAC+Eprior+Eregional(*)
wherein E isGACEnergy of geometric active contour line model, EpriorBeing the shape prior energy, EregionalIs the regional energy.
Specifically, the energy calculation formula of the geometric active contour line model is as follows:
Figure BDA0003224128710000053
where α, λ are the weights of the two energy terms, Φ is the level set function, H is the Heaviside function, Ω is the image region, δ is the Dirac function,
Figure BDA0003224128710000051
is a gradient operator, which is a linear operator,
Figure BDA0003224128710000052
is the gradient of phi; g is a boundary operator, toolThe body expression is:
Figure BDA0003224128710000061
wherein I is the gray scale of the image, GσIs a gaussian kernel function.
The shape prior energy calculation formula is as follows:
Figure BDA0003224128710000062
where β is the weight of the shape prior energy,
Figure BDA0003224128710000063
is a level set function of shape priors; r, c are shape prior parameters, r is radius, c is centerline;
the formula for calculating the area energy is as follows:
Eregional＝γ∫Ω(I-f1)2H(Φ)+(I-f2)2(1-H(Φ))dx
where γ is the weight of the region energy, I is the gray level of the image, f1，f2The average values of the image gradations of the regions where H is 1 and H is 0, respectively.
And (4) solving the movable contour model to obtain the segmentation result of the image.
In the three parts of the total energy functional, one is an expression of a classical GAC method, mainly related to the boundary information of the image, one is an expression related to shape prior, and the other is an expression related to region information; when a specific image and a segmentation curve are brought into the calculation, a value is obtained, and the value is different when the segmentation curve is different. The energy functional is designed as follows: when the segmentation curve reaches the target boundary, the energy E will be lowest, so solving the energy functional is to find where the minima of this energy functional occur. Because the energy functional is complex, the minimum is found by one-step iterative computation, and a new position of the segmentation curve can be found by each step of iterative computation. The evolution of the primary curve can also be said to have reached a new position.
Solving the active contour model refers to solving the minimum of the energy functional, updating the shape prior according to the obtained segmentation curve, and circularly performing the solving-updating operation until a stopping condition is reached to obtain a segmentation result. Here, "solving the functional" is a process of solving the active contour line so that the functional becomes extremely small. "segmentation" means the division of an image into two parts. As explained above, the position of the segmentation curve determines the magnitude of the E energy, which is calculated as the regional energyregionalFor example, f1 and f2 are the gray levels of the regions where H is 0 and H is 1, respectively, and the value of H is determined by the dividing curve, specifically, the inner region H of the curve is 1 and the outer region is 0. Therefore, changes in the position of the segmentation curve will result in changes in these energy functional values such that the energy functional reaches the position of the extremely small segmentation curve, i.e., the intended target boundary.
The invention adopts a fast algorithm to solve the energy functional, the fast algorithm adopts a characteristic function to represent the divided areas, and the characteristic function is as follows:
Figure BDA0003224128710000071
where Γ is the segmentation curve of the image, ΩΓAn inner region representing a curve Γ;
then, the boundary integral for Γ is approximated by a feature function u as:
Figure BDA0003224128710000072
or
Figure BDA0003224128710000073
Wherein, denotes the convolution calculation, GτExpressed as:
Figure BDA0003224128710000074
then, the energy functional (×) approximation is expressed in the form:
Figure BDA0003224128710000075
wherein the content of the first and second substances,
Figure BDA0003224128710000076
is a shape prior represented by a characteristic function;
find out Eτ(u) to a minimum of u, i.e. an optimal segmentation result.
In the embodiment of the present invention, the following calculation is providedτ(u) method to achieve minimal u:
will Eτ(u) linearization, i.e. u at the kth iterationkIs subjected to a first order Taylor expansion, Eτ(u) is approximated as follows:
Lτ(u,uk)＝∫Ωuψkdx
wherein the content of the first and second substances,
Figure BDA0003224128710000077
then u for the k +1 th iterationk+1Obtained by solving the following linear problem:
Figure BDA0003224128710000081
since the minima of the linear functional on the convex set must be reached at the boundary, the following solution is:
Figure BDA0003224128710000082
in summary, the solution process of the active contour model in the embodiment of the present invention is as follows:
(a) fast solution of curve evolution, calculating psik：
Figure BDA0003224128710000083
For psikPerforming threshold processing to obtain:
Figure BDA0003224128710000084
wherein k is 0,1,2 … …, u0Namely the initialized cylindrical surface obtained in the step (2).
(b) Determining a new shape prior from the evolution curve:
the three-dimensional segmentation result obtained in step (a) is recorded as ujBuild a new listj. According to ujFor the segmentation curve on each frame of the three-dimensional OCT image
Figure BDA0003224128710000085
Updating the shape prior, specifically: in each of the frames, the frame number is determined,
Figure BDA0003224128710000086
the center of the represented area is used as a new circle center
Figure BDA0003224128710000087
Figure BDA0003224128710000088
The radius of the minimum circumscribed circle of (a) is taken as a new radius, and a new circular area is generated based on the new radius
Figure BDA0003224128710000089
Where i is 1,2 … n, which is the number of frames of the three-dimensional image.
Circular area to be generated
Figure BDA00032241287100000810
Comparing with the reference image if
Figure BDA00032241287100000811
Is disjoint with the reference image of the corresponding frame, then is included in the listj. Calculating listjThe average value of all radii in (1) is recorded as rjThen the new shape is a priori
Figure BDA00032241287100000812
Then from the center line
Figure BDA00032241287100000813
And radius rjAnd (4) uniquely determining.
(c) Repeating steps (a) - (b) until uj+1＝ujOr listjIs empty.
The following is a specific example to verify the effect of the method of the present invention.
Aiming at 12 three-dimensional OCT coronary vessel images, 3240 frames of two-dimensional images are counted, the Image segmentation method based on the self-adaptive shape prior is compared with the segmentation result of a segmentation method without prior and an interpolation model (see the detailed results of a Lambrs Athanasiou et al full automatic segmentation and interpolation optical coherence tomography. in: Medical Imaging 2017: Image processing. Vol.10133.International Society for Optics and photometics.2017, p.101233I.), and the segmentation performance is compared by adopting a Dice coefficient, and the calculation formula is as follows:
Figure BDA0003224128710000091
where A is the segmentation result and G is the actual target region manually labeled by the physician. The coincidence degree between the segmentation result and the real target region is measured by the Dice, and the experimental method is more effective when the Dice is higher. In the experiment, the parameters α ═ 1, β ═ -0.5, λ ═ 0.2, and γ ═ 1 were taken. Table 1 shows the comparison of the Dice values obtained by applying different segmentation models to 12 three-dimensional OCT images.
Table 1 example Dice values for each model segmentation
Model (model) Dice
Interpolation model 86.7±7.3
Shape-free prior model 75.0±4.8
Model of the invention 93.6±2.4
Experiments show that when the three-dimensional OCT image is segmented by the model, the segmentation result is superior to that of other two model methods.
FIG. 5 is a comparison of segmentation results of partially less disturbed images according to different models, FIG. 6 is a comparison of segmentation results of partially more disturbed images according to different models, wherein the first column is an initial Image, the second column is a manually labeled real target boundary, the third column is an interpolation model (see in particular "Lambar S analysis et al. full automatic segmentation of interferometric coherent coherence tomography. in: Medical Imaging 2017: Image processing. Vol.10133.International Society for Optics and photometics.2017, p.101332I."), the fourth column is a segmentation result of the inventive model with shape prior removed, and the fifth column is a segmentation result of the inventive model. It can be seen that the results obtained using the model segmentation of the present invention are significantly better than other methods, especially in the more disturbed image shown in fig. 6.
According to another embodiment of the present invention, there is provided a medical imaging apparatus including: memory storing one or more computer programs which, when executed by one or more processors, cause the one or more processors to perform steps comprising the fully automated medical image segmentation method based on shape priors as described above.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.

Claims (9)

1. A full-automatic medical image segmentation method based on shape prior is characterized by comprising the following steps:
(1) selecting a three-dimensional OCT image to be processed, converting original OCT data into a rectangular coordinate system, and extracting a part with the maximum gray value as a reference image;
(2) detecting the outer wall of the catheter according to the three-dimensional OCT image, and generating the outer wall surface of the catheter;
(3) initializing the outer wall surface of the conduit as a curve and shape prior, and fusing the curve and the shape prior into a variational frame to obtain a geometric active contour model based on the shape prior;
(4) and solving the active contour model, updating the shape prior according to the obtained segmentation curve, and alternately performing the operation of solving the model and updating the shape prior until an iteration stop condition is reached to obtain a segmentation result.
2. The fully-automatic medical image segmentation method based on shape prior as claimed in claim 1, wherein the portion with the largest gray value is extracted as the reference image by two extra threshold values in step (1).
3. The fully-automatic shape-prior based medical image segmentation method according to claim 1, wherein the step (2) comprises: extracting two vertical sections which are vertical to each other of the three-dimensional OCT image, respectively calculating the gradient of each vertical section in the y-axis direction, and accumulating the gradient image in the x-axis direction; clustering the upper part and the lower part of the accumulated result respectively, and selecting the one with larger value from the two types, wherein the point farthest from the center of the image in the one type is the position of the outer wall of the catheter; obtaining two straight line positions in each longitudinal section image, and then obtaining four straight line positions in the two longitudinal sections; and correspondingly returning the linear coordinates to the three-dimensional image to obtain four straight lines in the three-dimensional image, and calculating the cylindrical surfaces of the four straight lines to obtain a standard cylindrical surface, namely the outer wall surface of the catheter.
4. The fully automatic shape-prior based medical image segmentation method according to claim 1, wherein the shape-prior based geometric activity contour model in step (3) is represented using the following total energy functional:
E＝EGAC+Eprior+Eregional (*)
wherein E isGACEnergy of geometric active contour line model, EpriorBeing the shape prior energy, EregionalIs the regional energy;
the energy calculation formula of the geometric active contour line model is as follows:
Figure FDA0003224128700000011
wherein, alpha and lambda are the weights of two energy terms, phi is a level set function, H is a Heaviside function, and g is a boundary operator; omega is the image area, delta is the Dirac function,
Figure FDA0003224128700000012
is a gradient operator, which is a linear operator,
Figure FDA0003224128700000013
is the gradient of phi;
the shape prior energy calculation formula is as follows:
Figure FDA0003224128700000021
where β is the weight of the shape prior energy,
Figure FDA0003224128700000022
is the level set function of the shape prior, r, c are the parameters of the shape prior, r is the radius, c is the centerline;
the calculation formula of the region energy is as follows:
Eregional＝γ∫Ω(I-f1)2H(Φ)+(I-f2)2(1-H(Φ))dx
where γ is the weight of the region energy, I is the gray level of the image, f1，f2The average values of the image gradations of the regions where H is 1 and H is 0, respectively.
5. The method of claim 4, wherein the boundary operator is specifically expressed as:
Figure FDA0003224128700000023
wherein G isσIs a gaussian kernel function.
6. The fully-automatic shape-prior-based medical image segmentation method according to claim 4, wherein in the step (4), solving the active contour model comprises: solving the total energy functional by using a fast algorithm, wherein the fast algorithm represents the segmented region by using a characteristic function, and the characteristic function is as follows:
Figure FDA0003224128700000024
where Γ is the segmentation curve of the image, ΩΓAn inner region representing a curve Γ;
the boundary integral for Γ is approximately represented by a characteristic function u as:
Figure FDA0003224128700000025
or
Figure FDA0003224128700000026
Wherein denotes a convolution calculation, n denotes a spatial dimension, RnRepresenting a space of n dimensions, τ being a parameter with respect to G, GτExpressed as:
Figure FDA0003224128700000031
the total energy functional (×) is then approximately represented in the form:
Figure FDA0003224128700000032
wherein the content of the first and second substances,
Figure FDA0003224128700000033
is a shape prior represented by a characteristic function;
find out Eτ(u) to a minimum of u, i.e. an optimal segmentation result.
7. A method of fully automatic medical image segmentation based on shape priors as claimed in claim 6 characterized in that the solution E is derivedτ(u) to a minimum u includes: will Eτ(u) linearization, i.e. u at the kth iterationkIs subjected to a first order Taylor expansion, Eτ(u) is approximated as follows:
Lτ(u，uk)＝∫Ωuψkdx
wherein the content of the first and second substances,
Figure FDA0003224128700000034
then u for the k +1 th iterationk+1Obtained by solving the following linear problem:
Figure FDA0003224128700000035
since the minima of the linear functional on the convex set must be reached at the boundary, the following solution is:
Figure FDA0003224128700000036
8. the fully-automatic shape-prior-based medical image segmentation method of claim 6, wherein the updating of the shape prior according to the obtained segmentation curve comprises: defining the height of all cylinders as a pixel, determining a central line and a radius according to the position of a segmentation curve on each frame in an image, determining cylinders according to the central line and the radius, comparing each newly generated cylinder with a reference image, recording the radius of the cylinder which does not intersect with the reference image, and calculating the average value of the radius of the cylinders to serve as the uniform radius of the shape prior.
9. A medical imaging device, comprising:
memory storing one or more computer programs which, when executed by one or more processors, cause the one or more processors to perform the steps of the fully automated shape-prior based medical image segmentation method of any one of claims 1-8.
CN202110966535.4A 2021-08-23 2021-08-23 Full-automatic medical image segmentation method and device based on shape prior Pending CN113706494A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110966535.4A CN113706494A (en) 2021-08-23 2021-08-23 Full-automatic medical image segmentation method and device based on shape prior

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110966535.4A CN113706494A (en) 2021-08-23 2021-08-23 Full-automatic medical image segmentation method and device based on shape prior

Publications (1)

Publication Number Publication Date
CN113706494A true CN113706494A (en) 2021-11-26

Family

ID=78653947

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110966535.4A Pending CN113706494A (en) 2021-08-23 2021-08-23 Full-automatic medical image segmentation method and device based on shape prior

Country Status (1)

Country Link
CN (1) CN113706494A (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070014473A1 (en) * 2005-07-15 2007-01-18 Siemens Corporate Research Inc System and method for graph cuts image segmentation using a shape prior
US20080075375A1 (en) * 2006-07-24 2008-03-27 Siemens Corporate Research, Inc. System and Method For Statistical Shape Model Based Segmentation of Intravascular Ultrasound and Optical Coherence Tomography Images
US20170053417A1 (en) * 2015-08-21 2017-02-23 The Regents Of The University Of California Automated comparison of 3d images
CN107240114A (en) * 2017-05-27 2017-10-10 南京林业大学 A kind of semi-automatic medical image cutting method based on a distance function shape constraining
EP3591614A1 (en) * 2018-07-06 2020-01-08 Charité - Universitätsmedizin Berlin Method and computer program for segmentation of optical coherence tomography images of the retina
CN112862834A (en) * 2021-01-14 2021-05-28 江南大学 Image segmentation method based on visual salient region and active contour

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070014473A1 (en) * 2005-07-15 2007-01-18 Siemens Corporate Research Inc System and method for graph cuts image segmentation using a shape prior
US20080075375A1 (en) * 2006-07-24 2008-03-27 Siemens Corporate Research, Inc. System and Method For Statistical Shape Model Based Segmentation of Intravascular Ultrasound and Optical Coherence Tomography Images
US20170053417A1 (en) * 2015-08-21 2017-02-23 The Regents Of The University Of California Automated comparison of 3d images
CN107240114A (en) * 2017-05-27 2017-10-10 南京林业大学 A kind of semi-automatic medical image cutting method based on a distance function shape constraining
EP3591614A1 (en) * 2018-07-06 2020-01-08 Charité - Universitätsmedizin Berlin Method and computer program for segmentation of optical coherence tomography images of the retina
CN112862834A (en) * 2021-01-14 2021-05-28 江南大学 Image segmentation method based on visual salient region and active contour

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
A. ELTANBOLY等: ""A novel automatic segmentation of healthy and diseased retinal layers from OCT scans"", 2016 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP), 19 August 2016 (2016-08-19), pages 2381 - 8549 *
贵鹿颖: ""基于变分偏微分方程的医学图像病灶的检测和分割"", CNKI博士学位论文全文库信息科技辑, no. 7, 15 July 2019 (2019-07-15), pages 43 - 44 *

Similar Documents

Publication Publication Date Title
US7773792B2 (en) Method for segmentation of IVUS image sequences
CN111951277B (en) Coronary artery segmentation method based on CTA image
CN110490040B (en) Method for identifying local vascular stenosis degree in DSA coronary artery image
CN107563983A (en) Image processing method and medical imaging devices
Chen et al. Kinematic and deformation analysis of 4-D coronary arterial trees reconstructed from cine angiograms
Xu et al. An improved algorithm for vessel centerline tracking in coronary angiograms
EP2357609B1 (en) Adaptative hit-or-miss region growing for vessel segmentation in medical imaging
CN111047607B (en) Method for automatically segmenting coronary artery
CN112308846B (en) Blood vessel segmentation method and device and electronic equipment
CN114693710A (en) Blood vessel lumen intimal contour extraction method and device, ultrasonic equipment and storage medium
CN112950537A (en) Coronary blood flow reserve fraction acquisition system, method and medium
Hammouche et al. Automatic IVUS lumen segmentation using a 3D adaptive helix model
CN116051738A (en) Method for reconstructing coronary artery blood vessel model based on CTA image and readable storage medium
Cao et al. Automatic identification of side branch and main vascular measurements in intravascular optical coherence tomography images
CN114998292A (en) Cardiovascular calcified plaque detection system based on residual double attention mechanism
CN117036530B (en) Cross-modal data-based coronary artery fractional flow reserve prediction method and device
CN112651984A (en) Blood vessel lumen intimal contour extraction method and device, ultrasonic equipment and storage medium
CN113706494A (en) Full-automatic medical image segmentation method and device based on shape prior
Zamani Boroujeni et al. Coronary artery center-line extraction using second order local features
Khaled et al. Automatic fuzzy-based hybrid approach for segmentation and centerline extraction of main coronary arteries
KR101484051B1 (en) Apparatus and method for preprocessing for CAD system using active contour method
Moshfegh et al. Development of an innovative technology to segment luminal borders of intravascular ultrasound image sequences in a fully automated manner
Ma et al. A novel automatic coronary artery segmentation method based on region growing with annular and spherical sector partition
CN116934885A (en) Lung segmentation method, device, electronic equipment and storage medium
Gui et al. Shape prior generation and geodesic active contour interactive iterating algorithm (SPACIAL): fully automatic segmentation for 3D lumen in intravascular optical coherence tomography images

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination