CN108537817B - Motion estimation method based on multi-scale spherical enhancement filter and level set algorithm - Google Patents

Abstract

The invention discloses a motion estimation method based on multi-scale spherical enhancement filter and level set algorithm, which relates to the problem of liver tumor tracking of ultrasonic free breathing sequence. It is characterized in that it effectively solves the problems of instability and poor real-time performance of the liver tumor motion tracking method under free breathing. The steps of the present invention are as follows: step 1, preprocessing the obtained ultrasound image sequence; step 2, using a multi-scale spherical enhancement filter to perform rough edge extraction on the target area; step 3, based on the rough edge extraction of the target area, using the The CV model of the level set algorithm extracts clear boundary information; Step 4: Determine the geometric center of the target and update the search area. The invention uses multi-scale spherical enhancement filter and CV model based on level set algorithm to process existing ultrasound image sequence, extracts clear boundary information of target area, is suitable for free breathing sequence, and ensures the stability and stability of liver tumor motion tracking process. real-time.

Description

Motion Estimation Method Based on Multiscale Spherical Enhancement Filter and Level Set Algorithm

technical field

The invention relates to medical ultrasonic imaging technology, in particular to the problem of liver tumor tracking of ultrasonic free breathing sequences, and is a motion estimation method based on a multi-scale spherical enhancement filter and a level set algorithm.

Background technique

Liver tumors refer to tumor lesions that occur in the liver, which can be classified into benign and malignant. Since the liver undertakes important metabolic functions of the human body, the occurrence of malignant tumors in the liver will lead to serious consequences. In addition, the liver is rich in blood supply, closely related to the important blood vessels of the human body, and the onset of malignant tumors of the liver is hidden and grows rapidly, so the treatment is very difficult.

Ultrasound imaging has been widely used in minimally invasive interventional therapy, especially for the treatment of human liver tumors, due to its high temporal resolution, high safety, and low side effects. However, the depth and position of the liver, and the motion caused by breathing make it difficult to accurately locate the lesion area. In clinical trials, in order to minimize the effects of breathing movements, methods of active breathing control, such as breath-holding and shallow breathing methods, are often used. However, these methods prolong treatment time and some patients cannot actively control their breathing. Therefore, how to continuously and stably track the movement of liver tumors in the free breathing state is particularly critical.

Some liver tumors are difficult to distinguish from healthy somatic cells due to their location and type, and it is extremely difficult to directly target these tumor cells for motion tracking. Therefore, simple and easily distinguishable anatomical landmarks, such as blood vessels and liver surface layers, are generally used to estimate the free breathing motion of the liver and indirectly predict tumor motion. Motion estimation methods are roughly divided into two categories: based on image grayscale and based on feature points. Among the methods based on image grayscale, the cross-correlation method and the absolute error weighting method are widely used. Because free breathing can lead to changes in the grayscale and shape of blood vessels in ultrasound images, and ultrasound images have strong speckle noise, motion estimation methods based on image grayscale are unstable and have large deviations. Specifically, the method based on feature points is to extract a series of feature points from the ultrasound image, match with the standard image to obtain the most similar area, calculate the spatial displacement vector, and complete the motion estimation. In general, feature point-based methods are more accurate than image grayscale-based methods. However, the extracted feature points are always suitable for a specific image sequence, and the process of feature point extraction and calculation requires a large amount of computation, which will prolong the tracking time. Therefore, how to find a suitable and simple feature point is of great significance to improve the accuracy and real-time performance of motion estimation.

Aiming at this problem, the present invention precisely locates the moving target based on the spherical enhancement filter and the level set algorithm. Due to the different acoustic parameters of human blood and tissue, the cross-section of hepatic blood vessels appears as a black spherical area in the ultrasound image, so it is easy to extract the rough outline of the target area using a multi-scale spherical enhancement filter. In order to avoid the intensity change in the blood vessel area, the boundary information is generally selected as the tracking target. The present invention uses the Chan-Vese (CV) model based on the level set algorithm to obtain the accurate boundary information of the moving target. In addition, the method of the present invention is tested using a clinical ultrasound free breathing sequence, and compared with the traditional cross-correlation method, the results further show the stability and accuracy of the method of the present invention.

SUMMARY OF THE INVENTION

The purpose of the present invention is to propose a motion estimation method of CV model based on multi-scale spherical enhancement filter and level set method, so as to improve the stability, accuracy and real-time performance of liver tumor motion tracking under free breathing.

The purpose of the present invention is achieved through the following technical solutions: using a multi-scale spherical enhancement filter to roughly extract the contour of the target area, then based on the roughly extracted contour, combined with the level set algorithm and the CV model, the boundary information of the target area is refined. Extraction, the geometric center of this boundary is the center of the target, and finally the search area is updated to continuously improve the found target area.

The flowchart of the present invention is shown in FIG. 1 , which is divided into four steps, and the specific steps are as follows.

Step 1: Preprocessing of the ultrasound image sequence.

Select a fixed large area in the liver ultrasound image in advance. This area should remove the liver capsule and some other organs, avoid blood vessels and rib artifacts as much as possible, and select a relatively uniform liver parenchyma. Because the ultrasound image has strong speckle noise, the liver ultrasound image obtained by preprocessing is selected by the mean filter with the window size set to 4*4.

Step 2: Roughly extract the edge of the target area.

Different from the traditional filter, the present invention proposes a Multi-scale Blobness Enhancement Filter (Multi-scale Blobness Enhancement Filter) in the process of extracting the rough outline of the target area.

The multi-scale refers to the selection of multiple Gaussian kernel values σ . The previous research on the selection of the Gaussian kernel value σ is to select a fixed value for convolution calculation. The idea of multi-scale Gaussian kernel value is proposed in the present invention. To extract the motion area of liver tumor under free breathing sequence, select multiple Gaussian kernel values for image convolution calculation.

1) The kernel area is selected by the standard position marked by the doctor in the liver ultrasound image, and the size of the kernel is positively related to the size of the target. Build a multiscale dataset of raw images.

，通过二者绝对值的比值将球状区域和线状、盘状区域区分开来，并且有效滤除背景中的斑点噪声，从而提取出肝血管横截面——较暗的球状区域。2) Perform the eigenvalue analysis of the Hessian matrix on each pixel in the ultrasound image, and extract the main direction of the local second-order structure of the decomposed image. The resulting eigenvalues are

, the spherical area is distinguished from the linear and disc-shaped areas by the ratio of the absolute values of the two, and the speckle noise in the background is effectively filtered out, so as to extract the hepatic blood vessel cross-section—the darker spherical area.

A schematic diagram of the effect of multi-scale processing and rough edge extraction is shown in Figure 2.

3) Strengthen the extracted spherical dark domain to make it more obvious in the processed graphics.

Step 3: Further extract clear boundary information based on the rough edge extraction of the target area.

Because the ultrasonic image has strong speckle noise, the traditional method of extracting boundary information based on gradient has poor reliability and clarity. The present invention uses the CV model based on level set algorithm to extract boundary information without any gradient information.

Minimize the energy function and replace the evolving closed contour with the level set function, processing the resulting equation using the Euler-Lagrangian method.

Step 4: Determine the geometric center of the target and update the search area.

1) Calculate the geometric center of the boundary obtained in step 2, which is the geometric center of the target area.

2) Calculate the size of the target area according to the geometric center and the boundary information in step 2. In order to suppress the influence of respiration, each side of the target area is extended accordingly to obtain a new target area.

3) Return to step 2 to recalculate the target area.

The invention can effectively suppress the influence of random low-echo regions on the tracking process, overcome the unreliable defect of boundary extraction in traditional methods, is well suited for free breathing sequences, ensures the stability of the tracking process, and meets clinical requirements.

The invention can also effectively distinguish the hepatic blood vessel cross-section from other linear, disc-shaped areas and background areas, and strengthen the spherical dark area of the blood vessel cross-section, effectively overcome the defect of strong speckle noise in ultrasonic images, and increase the accuracy of the tracking process.

Description of drawings

FIG. 1 is a flow chart of the present invention.

FIG. 2 is a schematic diagram of the effect of step 2 of the present invention.

Detailed ways

The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings, but are not limited thereto. Any modification or equivalent replacement of the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention shall be included in the present invention. within the scope of protection.

The present invention provides a motion estimation method based on a multi-scale spherical enhancement filter and a level set. The premise of realizing the method is the existence of an ultrasound liver image sequence with blood vessel cross-sections calibrated by medical experts under the condition of free breathing. Firstly, the image sequence is preprocessed to filter out the speckle noise of ultrasound, and reduce the loss of useful information as much as possible; then use the multi-scale spherical enhancement filter to roughly extract the contour of the target area. The CV model of the algorithm precisely extracts the boundary information of the target area, and finally updates the search area to continuously improve the found target area. The specific implementation steps are as follows.

Perform step 1, preprocessing of the ultrasound image sequence.

Mean filter: use m * m filter template, the center coordinate of the filter template is ( x , y ), and the gray value of the filtered image coordinate ( x , y ) is:

代表滤波器模板覆盖下的图像像素灰度值，m为滤波器模板的大小，以像素为单位计数。where f represents the image,

Represents the gray value of the image pixel covered by the filter template, m is the size of the filter template, counted in pixels.

Step 2: Use the multi-scale spherical enhancement filter to extract the rough outline of the target area.

1) The kernel area is selected by the standard position marked by the doctor in the liver ultrasound image, and the size of the kernel is positively related to the size of the target. A multi-scale data set of the original image is established. Based on the test data set of this patent, the maximum value of the scale σ is not more than 20 pixels, and the minimum value is not less than 3 pixels. In order to ensure real-time performance and reduce calculation time, 4 to 5 different scale information σ are selected.

2) Perform the eigenvalue analysis of the Hessian matrix on each pixel in the ultrasound image, and extract the main direction of the local second-order structure of the decomposed image. The Hessian matrix is represented as follows:

是超声图像的第i行，第j列对应像素位置的二阶差分；

是Hessian矩阵的特征值，通过二者绝对值的比值：Among them, taking a two-dimensional ultrasound image sequence as an example, σ is a vector with two components, and the method described in this patent is also applicable to a three-dimensional ultrasound image sequence. At this time, σ is a vector with three components, and the Hessian matrix can be further expressed It is in the form of a 3*3 order matrix;

is the second-order difference of the pixel position corresponding to the i -th row and the j -th column of the ultrasound image;

is the eigenvalue of the Hessian matrix, by the ratio of the absolute values of the two:

Distinguish spherical areas from linear and disc-shaped areas. At the same time, in order to effectively filter out the background, the following processing is performed:

Furthermore, the measurement of the spherical enhancement filter can be expressed as follows:

where α and β are the correlation coefficients. This results in the extraction of a cross-section of the hepatic vessels—the darker spherical region.

3) Strengthen the black spherical area to make it more obvious in the processed graphics.

Step 3: Use the CV model based on the level set algorithm to extract clear boundary information and minimize the energy function in the following formula:

是演化闭合轮廓的长度，I是待处理图像，

是C内外的平均灰度级，μ、

是权重系数，一般来说，

。in,

is the length of the evolving closed contour, I is the image to be processed,

is the average gray level inside and outside C , μ,

is the weight coefficient, in general,

.

To obtain the minimum value of the energy function, replace C with the level set function and use the Euler-Lagrange method to obtain the following equation:

是狄拉克函数，

，在本专利所使用的测试数据集中，

in

is the Dirac function,

, in the test data set used in this patent,

是在多尺度策略中得到的目标区域大致轮廓。

is the approximate outline of the target region obtained in the multi-scale strategy.

Step 4: Determine the geometric center of the target and update the search area.

1) Apply the accumulation method to calculate the geometric center of the target area using the boundary points obtained in step 2. Taking a two-dimensional image as an example, the formula for the cumulative sum method is as follows:

代表n个边界点中的第i个边界点。in,

represents the ith boundary point among the n boundary points.

2) Calculate the size of the target area according to the geometric center and the boundary information in step 2. In order to suppress the influence of respiration, each side of the target area is extended accordingly. In order to limit the area of the search area and improve the operation speed, generally 10-20 pixels can be extended, and the obtained area is the new target area.

3) Return to step 2 to recalculate the target area.

The present invention performs motion estimation based on multi-scale spherical enhancement filter and level set algorithm, is well suited for free breathing sequence, has good real-time performance of motion estimation, overcomes the disadvantage of unreliable boundary extraction of traditional methods, and effectively overcomes the problem of strong speckle noise in ultrasonic images. shortcomings, and to ensure the stability and accuracy of the tracking process to meet clinical requirements.

Claims (3)

1. a motion estimation method based on multi-scale spherical enhancement filter and level set algorithm, is characterized in that it comprises the following steps: Step 1, preprocessing of ultrasound image sequence; Step 2: Roughly extract the edge of the target area; the specific process is: 1) The kernel area is selected by the standard position marked by the doctor in the liver ultrasound image, and the kernel size σ is positively related to the size of the target; establish a multi-scale data set of the original image, in order to ensure real-time performance and reduce calculation time, select 4 to 5 different scale information; 2) Perform the eigenvalue analysis of the Hessian matrix on each pixel in the ultrasound image, and extract the main direction of the local second-order structure of the decomposed image. The eigenvalues of the Hessian matrix are λ ₁ , λ ₂ , through the ratio of the absolute values of the two:

Distinguish spherical areas from linear and disc-shaped areas; at the same time, in order to effectively filter out the background, the following processing is performed:

Furthermore, the measurement of the spherical enhancement filter can be expressed as follows:

Among them, α and β are the corresponding coefficients in the spherical enhancement filter; thus, the cross section of the hepatic blood vessel is extracted—the black spherical region; 3) Strengthen the black spherical area to make it more obvious in the processed graphics; Step 3: Based on the rough edge extraction of the target area, further extract clear boundary information; the specific process is as follows: Minimize the energy function in the formula:

Among them, L(C) is the length of the evolution closed contour, I is the image to be processed, c ₁ , c ₂ are the average gray levels inside and outside C, μ, γ ₁ , γ ₂ are the weight coefficients, γ ₁ =γ ₂ = 1; To obtain the minimum value of the energy function, replace C with the level set function and use the Euler-Lagrange method to obtain the following equation:

where δ(Ф) is the Dirac function, v=0, μ=0.1, u ₀ is the rough outline of the target region obtained in the multi-scale strategy; Step 4: Determine the geometric center of the target and update the search area. 2. The motion estimation method based on multi-scale spherical enhancement filter and level set algorithm according to claim 1, characterized in that in the step 1, the preprocessing method is: select a fixed hepatic ultrasound image in advance. If a large area is selected, a mean filter is used to preprocess the area to obtain a smoothed liver ultrasound image. 3. the motion estimation method based on multi-scale spherical enhancement filter and level set algorithm according to claim 1, is characterized in that described step 4 is: 1) Calculate the geometric center of the boundary obtained in step 2, and this center is the geometric center of the target area; 2) According to this geometric center and the boundary information in the execution step 2, calculate the size of the target area, in order to suppress the influence of respiration, carry out corresponding extension on each side of the target area, and the obtained area is the new target area; 3) Return to step 2 to recalculate the target area.

Images