JP4963278B2 - Boundary extraction by weighted least squares method using separation measure - Google Patents

Description

  The present invention relates to an image data boundary extraction method, program, and apparatus. In particular, the present invention relates to a method of extracting a lumen-plaque boundary (lumen surface) and a plaque-intima boundary (external elastic plate) in an intravascular ultrasound (IVUS) B-mode image.

  There is known a technique for obtaining a tomographic image (B-mode image) of a blood vessel by inserting a catheter having an ultrasonic probe at the tip into the blood vessel and rotating the ultrasonic probe (Patent Document). 1). FIG. 1 is a schematic diagram of intravascular ultrasound (IVUS), and FIG. 2 is an image display example of intravascular ultrasound (IVUS). Coronary artery disease can be diagnosed by intravascular ultrasound (IVUS), but there are “stable plaques” (stable plaques) and “unstable plaques” (thrombosis caused by rupture). These need to be diagnosed accurately. FIG. 3 is an explanatory diagram of stable and unstable plaques.

  In order to accurately diagnose a plaque in an intravascular ultrasound image, it is necessary to accurately determine the boundary between the lumen and the plaque and the boundary line between the plaque and the intima. FIG. 4 is a diagram illustrating a boundary line in an intravascular ultrasound image. In order to calculate an accurate area and volume, it is necessary to extract the boundary line with high accuracy, but at present, the boundary line is extracted manually by a doctor. FIG. 5 shows an example of conventional boundary line extraction. As shown in FIG. 5, conventionally, the user sets seed points on the boundary line while viewing the intravascular ultrasound image, and interpolates these seed points (for example, parametric spline interpolation), thereby defining the boundary line. I was asking. Since the boundary line is obtained by interpolation, there is a problem that an accurate boundary line cannot be extracted unless a large number of seed points are set, which takes a lot of time and effort and requires skilled knowledge and experience. In addition, methods that use image data to interpolate seed points include extraction methods using a dynamic contour model (SNAKE method) and extraction methods using a genetic algorithm. Processing is necessary and is not suitable for practical application. The snakes method (SNAKE method) is one of the methods for extracting the contour of the target region using the energy minimization principle. The energy function is defined on the contour line (SNAKE), and this energy function is minimized. The contour line is deformed so that the boundary, the line, the subjective contour line, and the like are obtained (Patent Document 2, Non-Patent Document 1).

In addition to this, there are Patent Documents 3 to 5 and Non-Patent Document 2 as conventional techniques related to the present invention. Non-Patent Document 2 regards an outline as “a region boundary that most separates an area from an area”, and represents an amount (separation degree) that represents the degree of separation of area feature amounts (luminance, hue, etc.) between the inside and the outside of the outline model. As a result, a method of extracting the contour of an object as “maximum degree of separation = boundary” has been proposed.
Special table 2003-503141 gazette "Intravascular ultrasonic analysis using effective contour line creation method and system" JP 7-128200 A "Monitoring Method and Apparatus" JP 2003-334194 A "Image processing apparatus and ultrasonic diagnostic apparatus" Japanese Patent Application Laid-Open No. 7-184892 “Ultrasonic Image Processing Apparatus and Method” Japanese Patent Application Laid-Open No. 7-093561 “Edge and Outline Extraction Device” M. Kass, A. Witik, and D. Terzopoulos: “Snakes: Active Contour Model”, International of Computer Vision, Vol. 1, No. 4, pp. 321-331 (1988) Information Processing Society of Japan Research Report “Computer Vision and Image Media” Vol.1994 No.27 (Document number: IPSJ-CVIM93088002, Document name: Object contour extraction based on the degree of separation between regions)

  In order to diagnose the tissue characteristics of intravascular plaques based on images obtained by intravascular ultrasound, it is important to extract plaque regions from images. The extraction was a lot of time and effort, and it was a tough work that required skilled knowledge and experience. According to the present invention, boundaries (especially vascular lumen contours and intimal contours) on image data (especially intravascular ultrasound image data) can be smoothed with fewer seed points (user set points) than before. It is another object of the present invention to provide a boundary extraction method, program, and apparatus for extracting at a higher speed.

In order to achieve the above object, the present invention has the following configuration.
A boundary extraction method for extracting a boundary on image data,
A temporary boundary setting step for obtaining a temporary boundary by interpolating between a plurality of seed points set by the user on the image data;
A separation measure distribution calculating step for calculating a separation measure at least for each pixel in the vicinity of the temporary boundary of the image data;
And a boundary extraction step of obtaining the boundary by polynomial approximation by a least square method using the separation measure in each pixel near the temporary boundary as a weighting coefficient.
A boundary extraction program for extracting boundaries on image data,
A temporary boundary setting step for obtaining a temporary boundary by interpolating between a plurality of seed points set by the user on the image data;
A separation measure distribution calculating step for calculating a separation measure at least for each pixel in the vicinity of the temporary boundary of the image data;
A boundary extraction program, comprising: a boundary extraction step for obtaining the boundary by polynomial approximation by a least square method using the separation measure at each pixel near the temporary boundary as a weighting coefficient.
A boundary extraction device for extracting a boundary on image data,
Temporary boundary setting means for obtaining a temporary boundary by interpolating between a plurality of seed points set by the user on the image data;
A separation measure distribution calculating means for calculating a separation measure at least for each pixel in the vicinity of the temporary boundary of the image data;
And a boundary extraction unit that obtains the boundary by polynomial approximation based on a least square method using the separation measure in each pixel near the temporary boundary as a weighting coefficient.

Moreover, you may have the following embodiments preferably.
In the separation measure distribution calculation step,
The separation measure at the pixel point (i, j) is to set the region A and the region B in the i direction or the j direction across the pixel point (i, j),
(Where n _A is the number of pixels in region A, n _B is the number of pixels in region B, μ _A is the average pixel value in region A, μ _B is the average pixel value in region B, and μ _{A + B} is pixel value average within the regions a and B, the set of pixels in I _a region a, a set of pixels in I _B region B, the pixel value of each pixel in the I pixel set)
Ask for.
The image data is a B-mode image;
The region A and the region B are set in the depth direction with the pixel point (i, j) in between.
The image data is obtained by expanding polar coordinate system image data into an orthogonal coordinate system in which one axis is an angle and the other axis is a distance from a center point.
The polar coordinate system image data is a B-mode image by intravascular ultrasound,
At least one of the lumen surface and the outer elastic plate in the blood vessel is extracted.

By selecting a small number of seed points, the contour (boundary) can be extracted by a computer more efficiently and accurately than in the past. In particular, the method of the present invention has the advantage that it can be easily applied to a real problem (real-time application) because the computational load on the computer is much smaller than before. Although this does not feel a great difference in processing on a two-dimensional image, processing of three-dimensional image data (cylindrical tube-like image) in which two-dimensional image data (blood vessel-like circular image) is laminated (for example, the long axis of a blood vessel) When it is used for extracting a boundary such as an inner wall of a blood vessel on a directional cross-sectional image, the user does not feel uncomfortable, leading to real-time processing, and this merit is enormous.
Since the feature information in the image data is used to extract the boundary line, the boundary extraction can be performed smoothly and accurately even with a small number of seed points. In addition, since the boundary line is obtained by using polynomial approximation by the least square method with the separation measure as a weighting factor, it is not necessary to perform repetitive operations such as the Snakes method or the genetic algorithm, and high-speed operations are possible. Furthermore, since it is only necessary to perform the calculation process in the vicinity of the temporary boundary obtained by interpolating the seed points set in advance, it is possible to effectively extract the boundary with a small amount of calculation.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 6 is a flowchart of the present invention. The region extraction method of the present invention includes the following procedures.
<Step 1> Determination of seed point on GUI <Step 2> First-order approximation by linear interpolation <Step 3> Calculation of separation measure <Step 4> Determination of approximation range by first-order approximation <Step 5> Weighted minimum two Each step will be described in detail below with polynomial approximation using multiplication.

<Step 1> Seed Point Determination on GUI Using a method similar to the conventional method described in FIG. 5, a seed point on the boundary line between the lumen and the plaque or between the plaque and the intima is set by a user's click.
The intravascular ultrasound image is polar coordinate system image data (upper diagram in FIG. 7) displaying ultrasound reflection data corresponding to the distance from the center point, and is a polynomial using the calculation of the separation measure and the weighted least square method In order to facilitate the approximation, image data (lower diagram in FIG. 7) developed in an orthogonal coordinate system in which the horizontal axis is the angle and the vertical axis is the distance from the center point will be used.

<Step 2> First-order Approximation by Linear Interpolation Next, a “temporary boundary line” is set by interpolating between the seed points set in Step 1. In this embodiment, for the sake of simplicity, linear interpolation is used for interpolation between seed points. FIG. 8 is an image in which temporary boundary lines are set by linear interpolation between seed points. As for the interpolation method between the seed points, any known method may be used, and spline interpolation or the like may be used.

<Step 3> Calculation of Separation Measure Next, a “separation measure” which is a local statistic at each pixel of the image data is calculated. The separation measure is also referred to as “in-class variance / inter-class variance ratio”, and is defined by the above equation (Equation 2). In other words, the separation measure is defined by the ratio of intraclass variance and interclass variance. The separation measure is also described in Non-Patent Document 2 above. FIG. 9 is an explanatory diagram of the separation measure. As shown in the figure, a region A and a region B are set across pixel points for which a separation measure is obtained, and a separation measure is calculated using the luminance values of the pixels in each region. As shown in the middle graph of FIG. 9, the separation measure becomes a large value when the difference between the luminance values of the region A and the region B is large. FIG. 10 is a graph obtained by calculating the separation measure distribution of the image data. FIG. 10 shows the separation measure calculated by setting region A and region B in the depth direction across the pixel point. As can be seen from FIG. 10, the separation measure has a high value at the boundary portion, but it is difficult to obtain the boundary line only from the separation measure distribution. FIG. 10 shows a color image displayed in gray, and the separation measure value of the dark portion (corresponding to red) surrounded by the portion having a medium separation measure (corresponding to yellow and green) is the highest.

<Step 4> Determination of Approximate Range by First-Order Approximation Next, a region near the temporary boundary obtained in Step 2 is set as an approximate range. This narrows down the calculation range in the next step. Since the boundary line to be obtained is in the vicinity of the temporary boundary line, it is possible to extract the boundary with higher accuracy by narrowing the focus range to the vicinity area of the temporary boundary line. As a method for setting the approximate range, any known method may be used. Further, in this embodiment, the separation measure distribution of the entire image data is obtained in Step 3, but since the separation measure needs to be calculated only in the vicinity of the temporary boundary line, by narrowing the focus range of the separation measure distribution, It is also possible to improve the calculation speed. FIG. 11 is a separation measure distribution in the approximate range set near the temporary boundary. In addition, the rectangular part (three places) on the temporary boundary line in FIG. 11 is a scale for clearly indicating the approximate range.

<Step 5> Polynomial Approximation Using Weighted Least Squares Method Next, a boundary line is extracted based on the separation measure distribution in which the calculation range is narrowed down in Step 4. FIG. 12 is an explanatory diagram of boundary line extraction by the weighted least square method. The boundary line is approximated by a P-th order polynomial, and each coefficient a _k of the polynomial F (j) is _obtained by the least square method. In the present invention, the temporary coefficient obtained in step 4 is used as a weighting coefficient in the least square method. A separation measure distribution S _ij in the vicinity of the boundary line is used. Thus, more accurate boundary line extraction can be performed with a relatively small amount of calculation.

  Below, the experimental result by the boundary extraction method of this invention is shown. FIG. 13 shows boundary extraction results ((a) conventional method, (b) present invention). X in the figure represents the seed point. FIG. 13 (a) uses conventional parametric spline interpolation, but it can be seen that this number of seed points is not practical because the boundary line is distorted. In order to accurately extract the boundary line by the conventional method, it is necessary to set more seed points. However, this requires more time and effort and requires skilled knowledge and experience. On the other hand, it can be seen that the boundary line can be accurately extracted even with a small number of seed points in the present invention of FIG.

Although an example of the embodiment of the present invention has been described above, the present invention is not limited to this, and it goes without saying that various modifications can be made within the scope of the technical idea described in the claims. Yes. For example, in the present embodiment, an example in which a boundary line on 2D image data is extracted has been described. However, since the separation measure distribution can be calculated even on 3D data, it can also be applied to extraction of boundary surfaces on 3D data. is there.

Schematic of intravascular ultrasound (IVUS). An image display example of intravascular ultrasound (IVUS). Explanatory drawing of stable plaque and unstable plaque. The figure which shows the boundary line in an intravascular ultrasonic image. Explanatory drawing of the setting of a seed point and the interpolation between seed points. The flowchart of this invention. Polar coordinate system display (upper figure) and orthogonal coordinate system display (lower figure) of blood vessel ultrasound image. An image in which a temporary boundary is set by linear interpolation between seed points. Explanatory drawing of a separation measure. The graph which calculated the separation measure distribution of image data. Separation measure distribution in the approximate range set near the temporary boundary. Explanatory drawing of the boundary line extraction by a weighted least square method. Boundary extraction results ((a) conventional method, (b) present invention).

Claims (7)

A boundary extraction method for extracting a boundary on image data,
A temporary boundary setting step for obtaining a temporary boundary by interpolating between a plurality of seed points set by the user on the image data;
A separation measure distribution calculating step for calculating a separation measure at least for each pixel in the vicinity of the temporary boundary of the image data;
And a boundary extraction step of obtaining the boundary by polynomial approximation by a least square method using the separation measure in each pixel near the temporary boundary as a weighting coefficient. In the separation measure distribution calculation step,
The separation measure at the pixel point (i, j) is to set the region A and the region B in the i direction or the j direction across the pixel point (i, j),
(Where n _A is the number of pixels in region A, n _B is the number of pixels in region B, μ _A is the average pixel value in region A, μ _B is the average pixel value in region B, and μ _{A + B} is pixel value average within the regions a and B, the set of pixels in I _a region a, a set of pixels in I _B region B, the pixel value of each pixel in the I pixel set)
The boundary extraction method according to claim 1, wherein the boundary extraction method is obtained by: The image data is a B-mode image;
3. The boundary extraction method according to claim 2, wherein the region A and the region B are set in the depth direction with the pixel point (i, j) interposed therebetween.   4. The boundary extraction method according to claim 3, wherein the image data is obtained by expanding polar coordinate system image data into an orthogonal coordinate system in which one axis is an angle and the other axis is a distance from a center point. The polar coordinate system image data is a B-mode image by intravascular ultrasound,
The boundary extraction method according to claim 4, wherein at least one of the lumen surface and the outer elastic plate in the blood vessel is extracted. A boundary extraction program for extracting boundaries on image data,
A temporary boundary setting step for obtaining a temporary boundary by interpolating between a plurality of seed points set by the user on the image data;
A separation measure distribution calculating step for calculating a separation measure at least for each pixel in the vicinity of the temporary boundary of the image data;
A boundary extraction program, comprising: a boundary extraction step for obtaining the boundary by polynomial approximation by a least square method using the separation measure at each pixel near the temporary boundary as a weighting coefficient. A boundary extraction device for extracting a boundary on image data,
Temporary boundary setting means for obtaining a temporary boundary by interpolating between a plurality of seed points set by the user on the image data;
A separation measure distribution calculating means for calculating a separation measure at least for each pixel in the vicinity of the temporary boundary of the image data;
And a boundary extraction unit that obtains the boundary by polynomial approximation based on a least square method using the separation measure in each pixel near the temporary boundary as a weighting coefficient.