KR20170000041A - Vessel Segmentation in Angiogram - Google Patents

Abstract

The present invention relates to a method for an automatic vessel segmentation in an angiogram, and more particularly, to a method for an automatic vessel segmentation in an angiogram which includes an augmenting step of augmenting a vessel, specially, based on a Hessian matrix, an extracting step of extracting a feature vector by using a unique value of the Hessian matrix to detect the thickness of a vessel, a learning step of mechanically learning a classifier with a database (DB) constructed based on the feature vector, and a testing step of testing the classifier. Accordingly, the present invention can accurately measure a stenosis degree and can improve a vessel detection probability.

Description

[0002] Vessel Segmentation in Angiography [0003]

The present invention relates to a technique for automatically segmenting blood vessels in an X-ray angiogram so that a catheter can be inserted in the correct direction.

Percutaneous coronary intervention (PCI) is a two-dimensional x-ray angiography (angiogram) procedure in which an operator inserts a catheter through the skin and places it in the coronary artery, ) And registration of the angiocentric images in three-dimensional CT can serve as a guide for guiding the catheter to be inserted in the correct direction, and it is possible to more precisely measure the degree of stenosis of the narrowed blood vessel .

However, during insertion of the catheter, since the contrasted blood vessel can not be confirmed in real time, it is entirely dependent on the feeling of the operator, and since the image obtained in the procedure is a projection image on the two-dimensional plane, Therefore, it is difficult to ensure the accuracy of the procedure, and there is a problem that the insertion mechanism of the body, such as a stent, may be erroneously inserted and removed.

A typical technique for the vascular segmentation in an X-ray angiogram is the technique of Frangi et al. And the technique of Krissian et al. In both of these techniques, In order to reflect the regional characteristics, we use the high value and eigenvector of Hessian matrix in multi scale.

However, Frangi et al. The technique defines the probability that a particular pixel is a vein using the ratio between the two norms of the two values and the two values, and Krissian et al. We use a response function that utilizes gradient information and gradient information in high resolution and eigenvectors.

Such Frangi et al. Technique and Krissian et al. Technique has a problem in that a false value is significantly increased due to the enhancement of the background noise rather than the blood vessel due to the simple use of simple numerical values such as the ratio and the size.

Korean Patent Publication No. 10-2015-0039641

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and it is an object of the present invention to provide a catheter for guiding a catheter to be inserted in a proper direction, more accurately measure the degree of stenosis of a blood vessel narrowed, It is an object of the present invention to provide a technique for automatically segmenting blood vessels in an angiogram that increases probability.

In order to achieve the above object, the technique of automatically segmenting blood vessels in an angiogram of the present invention is based on a Hessian matrix, and a Hessian matrix of blood vessels An analysis step of analyzing a two-dimensional image into the Hessian matrix to obtain a histogram, and an extraction step of extracting a feature vector using a Hessian matrix of the blood vessel obtained through the analysis step A learning step of generating a database based on the feature vector, a learning step of machine learning a random forest classifier into the database DB created in the generation step, A discriminating step for discriminating between large and small, a determining step for determining whether or not the blood vessel is not to be detected, and a zoning step for zoning the blood vessel.

In the analyzing step, the probability of each pixel being a blood vessel is determined by using a ratio between a vector size of a Hessian matrix of the blood vessel and a high value.

Extraction of the feature vector is used to vector the size of the scale σ _n eigenvalues and eigenvectors (λ _1, λ ₂₎ and the eigenvalues and eigenvectors ratio, the eigenvalues and eigenvectors (λ _1, λ ₂₎ of the (λ _1, λ ₂₎ of the And a feature vector is defined. A feature vector of the Hessian matrix is calculated by the number n of scales to define the entire feature vector.

And a testing step of testing the classifier after completing the learning step.

In the test step, the first feature vector of the first pixel of the input image is extracted and input to each of the individual non-leaf nodes of the random forest classifier. Then, all of the leaf nodes ) Of the blood vessel is determined based on the average value of the blood vessels.

And a computer-readable recording medium storing a program for causing a computer to perform a technique of automatically segmenting blood vessels in the angiogram.

According to the present invention, since the Hessian matrix is used, the blood vessel can be efficiently grasped by reflecting the narrow and long-range characteristic of the blood vessel.

By using the random forest classifier, it is possible to increase the probability of detecting a blood vessel, accurately measure the degree of narrowing of the blood vessel, and can effectively perform the procedure, reduce the positive error and increase the probability of the blood vessel, , The procedure time can be reduced, and the burden on both the doctor and the patient can be reduced.

In addition, since the catheter can be guided to be inserted in the correct direction, the safety of the procedure can be ensured, the risk can be reduced, and the degree of narrowing of the narrowed blood vessel can be accurately measured.

1 is a flow chart of a preferred embodiment of the present invention;
Figure 2 is a random forest algorithm diagram of Figure 1;
Fig. 3 is a comparison of the zoning result of Fig. 1 with other techniques. Fig.
4 shows a conventional vascular segmentation technique.

Preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a flow chart of a preferred embodiment of the present invention, FIG. 2 is a view of a random forest algorithm of FIG. 1, and FIG. 3 is a comparison of a zoning result of FIG. 1 and another technique.

Referring to FIG. 1, a technique of automatically segmenting blood vessels in an angiogram according to a preferred embodiment of the present invention includes a method of X-ray (X-ray) (S1) for analyzing an image in the Hessian matrix, and extracting a feature vector using the Hessian matrix of the blood vessel obtained through the analysis step (S1) (S5) of creating a database (DB) built on the basis of the feature vector; and a learning step (S5) of learning a random forest classifier in the database (DB) A determination step S7 for determining whether the feature vector of the database DB is large or small, a determination step S9 for determining whether the feature vector is a blood vessel or not, and a segmentation step S11 for segmenting the blood vessel .

The technique of automatically segmenting blood vessels in the angiogram of the present invention is based on a Hessian matrix. The Hessian matrix is suitable for optimizing reflecting the narrow and long-range characteristics of the blood vessel.

X-rays are taken of a part of the body.

An X-ray image is analyzed with the Hessian matrix to obtain a Hessian matrix of the blood vessel (S1, analysis step).

During the above-described analysis step (S1), when X-ray is taken, various organs including blood vessels, such as bones and diaphragm, are present, in order to view only blood vessels.

That is, it filters the blood vessel based on a Hessian matrix matrix. The Hessian matrix for each pixel in an X-ray image can be said to determine the likelihood that each pixel is a blood vessel through analysis of high values (λ ₁ , λ ₂ ). The likelihood of each pixel is determined by the vector magnitudes of the first and _second magnitudes λ ₁ and λ ₂ and the vector magnitudes of the first magnitude λ ₁ and the second magnitude λ ₂ , To be determined.

Based on the high-level analysis, the following equation (1) is calculated to calculate the vesselness value, which means likelihood that the pixel p is a blood vessel.

[Equation 1]

In this case, β and γ mean the thresholds that affect the sensitivity of R _β ² and S, respectively, and V _Frangi means Frangi et al. And R _β ² means the _ratio of the first high value (λ ₁ ) to the second high value (λ ₂ ). S is a second order structureness value that increases or decreases according to the structure, and calculates a vector size of the Hessian matrix according to a vector norm calculation method in the Frobenius matrix .

In order to detect blood vessels of various thicknesses (thicknesses), the vesselness value v at a specific scale? In the above equation (1) can be expanded to multiscale, as shown in equation (2).

&Quot; (2) "

와 그에 상응하는 스케일을 출력함으로써, 다양한 두께의 혈관을 검출할 수 있게 되는 것이다. 즉, 스케일 σ이 클수록 두꺼운 혈관을 검출할 수 있게 되는 것이다.The largest vesselness value per pixel

And the scale corresponding thereto, it is possible to detect blood vessels of various thicknesses. That is, the larger the scale?, The thicker the blood vessel can be detected.

In this case, v (p) means vesselness value at the particular pixel p and, σ means the scale value and, σ _min means the minimum value of the scale, and, σ _max means the maximum value of the scale, max _{v (p, σ)} means the largest vesselness value and corresponding scale.

A feature vector is extracted using a Hessian matrix of the blood vessel obtained through the analysis step (S1) (S3, extraction step)

In the extraction step (S3), feature vectors are extracted so as to detect blood vessels having various thicknesses.

The feature vector is extracted by combining the high-frequency values at various scales.

The feature vector, the first scale example example the eigenvalues and eigenvectors analysis in (σ _1), the first eigenvalues and eigenvectors (λ _1), the second eigenvalues and eigenvectors (λ ₂₎ and the first eigenvalues and eigenvectors (λ ₁₎ and ( ₃ ) using the ratio of the second high value (? ₂ ) and the vector magnitude of the first high value (? ₁ ) and the second high value (? ₂ ).

&Quot; (3) "

In this case, f _σ1 (p) means the feature vector of pixel p when the scale is σ ₁ (f is an abbreviation for feature), λ ₁ means the first high value, and λ ₂ means the second high value Max _σ v (p, σ) means the largest vesselness value and the corresponding scale, and min _σ v (p, σ) means the smallest vesselness value and corresponding scale.

The overall feature vector f (p) is calculated as follows using the Hessian matrix of various scales.

&Quot; (4) "

In this case, n denotes the number of scales to be used, and f (p) denotes a feature vector of pixel p when the scale is multi-scale.

That is, the total feature vector is defined by calculating a high value (? _N ) of the Hessian matrix by the number of the scales (? _N ).

(DB) based on the entire feature vector (S5, generation step)

The classifier is learned based on the feature vector constructed in the database DB (S5)

The random forest classifier is machine-learned in the database (DB) created in the generation step (S5) (S7, learning step)

The learning step S7 may be performed only at the initial stage of applying the random forest algorithm, storing the learning results, and thereafter using the vascular segmentation technique without going through the learning step S7 based on previously stored learning results Do.

After learning the random forest classifier, the learned classifier is tested.

In the test step, the feature vector f (p) of the pixel p of the input image is extracted and input to the non-leaf nodes of the already learned random forest classifier, and then all the leaf nodes the average value of the leaf nodes has a probability that the corresponding pixel belongs to the blood vessel.

The feature vector of the database DB is determined to be large or small (S9, discrimination step)

In the individual tree learning constituting the random forest classifier, training data randomly extracted from the database (DB) constructed beforehand is used. The decision at each node in the random forest classifier is defined as a thresholding of one of the elements of the feature vector f (p).

까지의 거리로 나타냈다. 즉, 각 노드(non-leaf node)에서는 대소판별을 할 때의 임계값(threshold)을 학습한다고 할 수 있는데, 이 과정을 통해 혈관인지 아닌지를 판별할 수 있는 것이다.As described in (5), in the present invention, a threshold value is calculated from the i-th element of the entire feature vector f (p)

Respectively. That is, in a non-leaf node, it can be said that a threshold value for discriminating between large and small is learned. Through this process, it is possible to determine whether or not it is a vessel.

That is, after inputting the feature vector f (p) to each of the non-leaf nodes, an average value of the result values of the generated non-leaf nodes is obtained. This average value determines the probability of the blood vessel. (S11, decision step)

&Quot; (5) "

In this case, θ denotes a threshold value, and θ _i ^j denotes a threshold value of the distance from the i-th element to the j-th element.

As shown in FIG. 2, the random forest is automatically applied to the vascular segmentation technique in the angiogram of the present invention.

When it is determined whether or not the blood vessel is determined (S11), the determined blood vessel is searched (S13)

In order to evaluate the performance of the automatic vessel segmentation algorithm using the random forest classifier in the present invention, the random forest algorithm and the Frangi et al. Technique and Krissian dt al. I compared the results of the automatic segmentation technique.

The results obtained from each technique are compared with the manual segmentation sensitivity and DSC (Dice Similarity Coefficient) as shown in [Table 1]. Compared to the existing two methods, the technique using the random forest algorithm Table 1 shows that a significant decrease in false positives is observed.

Random forest algorithm Frangi et al. Krissian et al. Sensitivity (%) 67 29 39 DSC (%) 56 18 23

In addition, as shown in FIG. 3, when comparing the conventional technique with the automatic technique of the arterial segmentation using the random forest algorithm, the false positive ) Is significantly decreased.

The technique of automatically segmenting the blood vessels in a two-dimensional X-ray angiogram using the random forest algorithm allows the practitioner to more accurately grasp the degree of vessel stenosis, And it is advantageous that the accuracy of the procedure can be improved by inducing the insertion of the catheter in the correct direction through the matching.

A computer-readable recording medium storing a program for causing a computer to execute the above-described method, and a detailed description of a preferred embodiment of the present invention will be given below with reference to the accompanying drawings. It should be noted that the same configurations in the drawings denote the same reference numerals whenever possible. Specific details are set forth in the following description, which is provided to provide a more thorough understanding of the present invention. In the following description of the present invention, detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

Throughout the specification, when a part is referred to as being "connected" to another part, it includes not only "directly connected" but also "electrically connected" with another part in between . Also, when an element is referred to as "comprising ", it means that it can include other elements as well, without departing from the other elements unless specifically stated otherwise.

May also be a hardware component, such as a processor or circuitry, and / or a software component, executed by a hardware component, such as a processor.

Meanwhile, the above-described embodiments of the present invention can be embodied in a general-purpose digital computer that can be created as a program that can be executed by a computer and operates the program using a computer-readable recording medium.

Computer readable media can be any available media that can be accessed by a computer and includes both volatile and nonvolatile media, removable and non-removable media. In addition, the computer-readable medium may include both computer storage media and communication media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Communication media typically includes any information delivery media, including computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave, or other transport mechanism.

It will be understood by those skilled in the art that the foregoing description of the present invention has been presented for illustrative purposes and that those skilled in the art will readily understand that various changes and modifications may be made without departing from the spirit or essential characteristics of the present invention. will be. It is therefore to be understood that the above-described embodiments are illustrative in all aspects and not restrictive. For example, each component described as a single entity may be distributed and implemented, and components described as being distributed may also be implemented in a combined form.

It is intended that the present invention covers the modifications and variations of this invention provided they come within the scope of the appended claims and their equivalents. .

As described above, the technique of automatically segmenting the blood vessels in the angiogram according to the present invention is particularly effective in the case of percutaneous coronary intervention (PCI), such as a two-dimensional X-ray angiogram angiogram) is suitable for the technique of automatically segmenting blood vessels.

Claims (6)

The present invention relates to a technique for automatically segmenting blood vessels in an angiogram based on a Hessian matrix,
An analysis step of analyzing a two-dimensional image into the Hessian matrix to obtain a Hessian matrix of blood vessels;
An extraction step of extracting a feature vector using a Hessian matrix of the blood vessel obtained through the analysis step;
Generating a database (DB) based on the feature vector;
A learning step of mechanically learning a random forest classifier in the database (DB) created in the generating step;
A discriminating step of discriminating the feature vector of the database (DB) from a large one;
A determination step of determining whether or not the blood vessel is a blood vessel;
A method for automatically segmenting blood vessels in an angiogram comprising a segmentation step of segmenting blood vessels. The method according to claim 1,
In the analysis step, an angiogram automatically determines the probability that each pixel is a blood vessel, using a ratio between a vector size of the Hessian matrix of the blood vessel and a high value. Technique for regionalization of blood vessels. The method of claim 2,
Using the vector magnitude of the eigenvalues and eigenvectors (λ _1, λ ₂₎ and, the eigenvalues and eigenvectors ratio, the eigenvalues and eigenvectors (λ _1, λ ₂₎ of the (λ _1, λ ₂₎ in the feature vector, the scale σ _n , ≪ / RTI >
Wherein a total feature vector is defined by calculating a height value of the Hessian matrix by a number n of the scales. The method of automatically segmenting an artery in an angiogram. The method of claim 3,
A technique for automatically segmenting blood vessels in an angiogram, further comprising a test step after the learning step and testing the classifier. The method of claim 4,
In the test step, the first feature vector of the first pixel of the input image is extracted and input to each of the individual non-leaf nodes of the random forest classifier. Then, all of the leaf nodes ) Of the blood vessel is determined based on the average value of the angiograms. A computer-readable recording medium storing a program for causing a computer to execute the method according to any one of claims 1 to 5.

Images