WO1994018639A1

WO1994018639A1 - Method and apparatus for generating well-registered subtraction images for digital subtraction angiography

Info

Publication number: WO1994018639A1
Application number: PCT/US1994/001106
Authority: WO
Inventors: Ping Hua
Original assignee: Siemens Medical Systems, Inc.
Priority date: 1993-02-09
Filing date: 1994-01-18
Publication date: 1994-08-18

Abstract

Contrast and mask images acquired for digital subtraction angiography are processed using an edge detection algorithm to produce edge images. Advantageously, a deterministic sign change algorithm is used to identify the mask edge image which is best registered with the contrast edge image.

Description

METHOD AND APPARATUS FOR GENERATING WELL-REGISTERED SUBTRACTION IMAGES FOR DIGITAL SUBTRACTION ANGIOGRAPHY

Background of the Invention

The invention relates to digital subtraction angiography (DSA) . More specifically, the invention relates to the computation of a digital subtraction image. In its most immediate sense, the invention relates to the identification of an optimal shift between a mask image and a contrast image, thereby facilitating production of a well-registered digital subtraction image.

In DSA, an X-ray image (herein, a "mask image") is acquired of a selected region of the body of a living patient. This image clearly shows anatomical features such as bones, body organs, etc but does not show the patient's vascular structure because this is transparent to X-rays.

Then, a contrast medium such as iodine is injected into the patient's bloodstream and another X-ray image (herein, a

"contrast image", also known as a "roadmap" or a "roadmap image") is acquired of the same region. Because the contrast medium is opaque to X-rays, the contrast image shows the vessels where bloodflow occurs. By subtracting one of the images from the other and thereby forming a subtraction image, the anatomical features are cancelled out and the vascular features are visually emphasized.

If the patient moves between acquisition of the mask and contrast images, the images will be misregistered with respect to each other and one of the images must be repositioned before the subtraction operation is carried out. Without such repositioning undesirable artifacts are created in the subtraction image. Such artifacts can diminish the diagnostic utility of the resulting subtraction image. Conventional DSA apparatus repositions images using pixel shift techniques. Such techniques are generally based on Fourier domain or image domain information. Both techniques are limited because corresponding pixels vary in intensity between the contrast image and the mask image. Additionally, because such techniques require substantial time and computer resources, it would be difficult to apply them in real time and they are conventionally carried out in a post-processing mode.

It would be advantageous to provide method and apparatus which could be easily implemented in hardware and therefore could, in real time, correct misregistration between the contrast and mask images.

One object of the invention is to provide method and apparatus for correcting, in real time and with limited computer and electronic resources, misregistration between the contrast and mask images.

Another object is, in general, to improve on prior art method and apparatus used in digital subtraction angiography.

In accordance with the preferred embodiment of the invention, the mask and contrast images are subjected to image processing to generate edge images. This emphasizes landmarks which can reliably be used to determine the misregistration between the contrast image and the mask image.

Advantageously, a region of interest ("ROI") is defined in the contrast image. (Advantageously, and in the preferred embodiment, this is done by defining the ROI in an initial subtraction image. This definition carries over into the contrast image. However, if the ROI is defined to be as large as the acquired image, no separate ROI definition step is required.)

Advantageously, and in accordance v/ith the preferred embodiment, the mask image and the contrast image are assumed to be misregistered by M pixels or less both horizontally and vertically; since the misregistration can be positive or negative in each direction, there are (2M+1)² ROI-shaped regions which are displaced by M pixels or less from the ROI. Then, a deterministic sign change (DSC) algorithm is applied (2M+1)² times; each time, the algorithm is applied to that part of the contrast edge image which is bounded by the ROI and that part of the mask edge image which is bounded by one of the ROI-shaped regions. In each instance, the resulting DSC score is recorded. The highest DSC score, and the single ROI-shaped region with which that highest DSC score is associated, are then identified. This identified ROI-shaped region is taken to encompass the best-registered mask image and the subtraction image is formed by subtracting between a) the contrast image bounded by the ROI and b) the mask image bounded by the identified ROI- shaped region.

Although in accordance with the preferred embodiment a DSC algorithm is applied and all of the (2M+1)² mask edge images bounded by these ROI-shaped regions are considered, this is only presently preferred. While use of a DSC algorithm is highly preferred because it can be easily implemented in hardware, another algorithm could be used instead, and it is also possible to implement the chosen algorithm in software if real time operation is not necessary. Additionally, it is anticipated that it will be possible to derive criteria by which such exhaustive comparisons with all the ROI-shaped regions are unnecessary and in accordance with which some of those images may be ignored. Thus, consideration of all of the mask edge images bounded by the ROI-shaped regions is not a part of the invention.

Brief Description of the Drawings

The invention will be better understood with reference to the following illustrative and non-limiting drawings, in which:

Fig. 1 is a schematic block diagram of DSA apparatus in accordance with the preferred embodiment of the invention;

Fig. 2 shows a conventional mask image;

Fig. 3 shows a conventional contrast image;

Fig. 4 shows a subtraction image formed by subtraction between the images in Figs. 2 and 3;

Fig. 5 shows a mask edge image derived from Fig. 2 in accordance with the preferred embodiment of the invention;

Fig. 6 shows a contrast edge image derived from Fig. 3 in accordance with the preferred embodiment of the invention;

Fig. 7 shows the relationship between the ROI and those ROI-shaped regions which are displaced by M pixels or less from the ROI; and

Fig. 8 is a flow chart showing in schematic fashion a method in accordance with the preferred embodiment of the invention. Detailed Description of a Preferred Embodiment

The herein-described preferred embodiment is considered especially appropriate for peripheral-artery and cerebral studies, for which most DSA studies are carried out. This is because for such studies, pixel shift techniques are adequate to correct for misregistrations between the mask and contrast images. However, this is not a part of the invention.

Additionally, and as is set forth below, the invention can be implemented either in hardware (e.g. a board installed in a DSA system which includes a computer) or in software which resides in the computer of a DSA system. The actual mode of implementation will be chosen in accordance with criteria such as operation speed, cost etc. For simplicity, the description below assumes that the invention is implemented in software resident in a computer.

A DSA unit includes an x-ray source 2, which directs a beam of x-rays through a patient 4. The transmitted X- rays strike an image intensifier 6, which projects an image onto a video camera 8. The video camera 8 then produces an output video signal which is recorded in, e.g. a disk drive 10 (or other storage device) in a computer (or processor) 12. An operator (not shown) of the computer causes a subtraction image of the ROI of the patient 4 to be displayed on a display device 14.

In conventional digital subtraction angiography, an X- ray mask image is taken of the patient 4. Such an image is shown in Fig. 2. Then, a contrast agent such as iodine is injected into the bloodstream of the patient 4 and a contrast image is acquired. Fig. 3 shows such a contrast image. A subtraction image is then computed by subtracting one of these images from the other one. Operation of conventional DSA apparatus requires an ROI to be defined in the contrast image. If desired in the preferred embodiment, this can be done directly, i.e. the operator may, using the computer 12, define the ROI with direct reference to the contrast image. However, the operator will usually wish to define the ROI with reference to a subtraction image, since a subtraction image emphasizes the vascular structure in which the diagnostician is interested. Thus, in the preferred embodiment, an initial subtraction image is formed by subtraction between the acquired mask and contrast images and this initial subtraction image is displayed to the user on display 14. (This initial subtraction image will likely contain artifacts because the acquired mask image will likely be misregistered with respect to the contrast image, but ROI definition does not require an artifact-free subtraction image.) Then, an ROI is defined in the initial subtraction image. Because any misregistration is taken to be in the mask image rather than in the contrast image, definition of an ROI in the initial subtraction image defines the identical ROI in the contrast image. For convenience, and in accordance with the preferred embodiment, the ROI is taken to be rectangular, but this is not required; the invention does not reside in the shape of the ROI.

Additionally, while definition of an ROI is presently preferred, it is not strictly necessary. This is because it would alternatively be possible to define, in advance, the ROI to be the same as the acquired contrast image or to encompass a predetermined region. In this instance ROI definition would not be a separate step but would be inherent in operation of the DSA apparatus. Thus, there is no need for the ROI to be of a different size than the contrast image. In conventional DSA apparatus, any misregistration between a mask image and a contrast image is determined using data in image3 such as are illustrated in Figs. 2 and 3. Such a determination requires substantial time and substantial resources because such images contain large quantities of data. (The resources may be computer resources where hardware signal processing is minimal or nonexistent; where hardware signal processing is used, resource utilization is determined by the specific design and configuration of the DSA system.) However, in accordance with the invention, this image data is processed to produce edge images.

Advantageously, and in accordance with the preferred embodiment, the image data is processed using the Mero- Vassy edge detection algorithm. This algorithm is chosen because it is simple, easily implementable in hardware and relatively insensitive to noise, but another algorithm could be used instead.

The edge images produced by this processing are shown in Figs. 4 and 5, which correspond to Figs. 2 and 3 respectively. These edge images enhance non-smooth texture in the original images, and also enhance the edges of bones and vessels.

It is assumed that the original mask and contrast images, and therefore the edge images shown in Figs. 4 and 5, are misregistered by M pixels or less in either direction along the horizontal and vertical axes. (M need not be integral and may indeed be less than 1; resolution on the subpixel order is possible and feasible using interpolation techniques.) For convenience and as stated above, it is assumed that the ROI is located in the correct location in the contrast image and that any misregistration is in the mask edge image. Since along an axis there always are 2M+1 pixels which are within M pixels of any particular pixel, and since there are two axes (horizontal and vertical), there are (2M+1)² ROI-shaped regions which are displaced by M pixels or less from the ROI (see Fig. 7.) Consequently, identification of the best-registered mask image is a matter of identifying that particular one of these (2M+1)² ROI-shaped regions which contains the best- registered mask image.

In accordance with the preferred embodiment of the invention, that portion of the contrast edge image which is bounded by the rectangular ROI is (see Fig. 6) individually compared with each portion of the mask edge image which is bounded by an ROI-shaped region that is displaced from the ROI by no more than M pixels in either direction along either or both of the horizontal and vertical axes. In further accordance with the preferred embodiment of the invention, each comparison is carried out using a deterministic sign change ("DSC") algorithm. In each case, this is done by creating an image D, wherein each pixel d(i,j) in the image D is constructed using the equation

d(i,j) = c(i,j) - m(i,j) + (-l)°^'+j>q

wherein c(i,j) is the pixel value at location (i,j) in the contrast edge image, m(i,j) is the pixel value at location (i,j) in the mask edge image and q is a parameter chosen in this instance to be 8. (The parameter q is related to the signal-to-noise ratio of the edge images, and must be chosen carefully. However, use of this DSC algorithm and this particular value of q is not a part of the invention and other algorithms and values can be chosen instead.) Then, each image D is scored by adding up the number of sign changes in it, i.e. the number of pairs of pixels, taken either horizontally or vertically, where the pixels in the pair are of different signs. The image D with the largest number of sign changes is then identified. The corresponding ROI-shaped region is then taken as bounding the best-registered mask image and the subtraction image is then formed using the contrast image bounded by the ROI and the mask image bounded by the corresponding ROI-shaped subregion.

Use of a DSC algorithm, and the above-disclosed particular DSC algorithm, is considered highly advantageous because it is easy to implement such an algorithm in hardware. However, it is possible to use other algorithms as long as they can suitably be applied to edge image data as described above.

In apparatus in accordance with the preferred embodiment of the invention, the computer 12 is programmed to operate in accordance with the flow chart shown in Fig. 8. Initially, contrast and mask images of the patient 4 are acquired. Then, in the usual (default) case, an initial subtraction image is formed by subtracting between the contrast and mask images. Thereafter, the operator defines an ROI in the initial subtraction image, using the computer 12. This defines the same ROI in the contrast image. (Alternatively, the operator can define the ROI in the contrast image directly.) In the preferred embodiment, the ROI is rectangular and is H x W pixels high by wide and its initial location will be assumed to be at position

Thereafter, the contrast and mask images are processed using the Mero-Vassy edge detection algorithm, thereby creating contrast edge and mask edge images.

Next, the computer is programmed to construct rectangles which have the same shape as the ROI and which are located within M pixels of it along either or both axes. Such ROI-shaped regions are located at (i+M,j+M) , (i+M,j-M), (i-M,j+M) , (i-M,j-M) , and every coordinate position in between (M being equal to the maximum misregistration, expressed in pixels, between the contrast image and the mask image.) There are (2M+1)² of such ROI- shaped regions. (Resolution at the pixel level can be achieved by pixel shifting. To achieve subpixel resolution, it is anticipated that the coordinates would be determined by a best-fit curve fitting scheme, and some sort of interpolation scheme would be used to perform subpixel subtraction.)

Thereafter, as has been described above, a DSC algorithm is applied to the contrast edge image bounded by the ROI and the mask edge image bounded by each ROI-shaped region and a DSC score is computed in each instance. This is carried out by the computer 12. The ROI-shaped region with the maximum DSC score is then identified. Finally, a best-registered subtraction image is created by subtracting between the contrast image bounded by the ROI and the mask image bounded by the ROI-shaped region with the maximum DSC score.

As stated above, it is not necessary that the above steps be carried out by a computer. It is alternatively possible to implement them in a hardware board or chassis which is connected to a computer. The exact mode of implementation of the invention is a matter of design choice.

Although a preferred embodiment has been described above, the scope of the invention is limited only by the following claims:

Claims

1. A method for generating a subtraction image in which the contrast image is well registered with the mask image, comprising the following steps: acquiring a mask image; acquiring a contrast image; defining, in the contrast image, a region of interest ("ROI") ; generating a mask edge image by applying an edge detection algorithm to the mask image; generating a contrast edge image by applying said edge detection algorithm to the contrast image; using a deterministic sign change (DSC) algorithm to compute, for a) each of a plurality of shifted mask edge images each being bounded by an ROI-shaped region and

b) that contrast edge image which is bounded by the region of interest a like plurality of DSC scores; identifying the highest DSC score so computed; and identifying that ROI-shaped region which corresponds to said highest DSC score.

2. The method of claim 1, further comprising the step of generating a subtraction image using the contrast image bounded by the ROI and the mask image bounded by said identified ROI-shaped region.

3. The method of claim 1, wherein said generating steps are carried out using a Mero-Vassy edge detection algorithm.

4. The method of claim 1, wherein the DSC algorithm is: creating an image D, wherein each pixel d(i,j) in the image D is constructed using the equation d(i,j) = c(i,j) - m(i,j) + (-l)<^i+J)q wherein c(i,j) is the pixel value at location (i,j) in the contrast edge image, m(i,j) is the pixel value at location (i,j) in the mask edge image and q equals 8; and the DSC score equals the number of sign changes in the image D.

5. The method of claim 1, wherein said plurality of shifted mask edge images is equal to or less than (2M+1)², wherein M equals an assumed maximum misregistration of the acquired images, expressed in pixels.

6. The method of claim 1, wherein said defining step comprises the steps of generating an initial subtraction image by subtracting between the contrast and mask images and defining an ROI in said initial subtraction image.

7. A method for generating a subtraction image, comprising the following steps: acquiring a mask image; acquiring a contrast image; generating a mask edge image by applying an edge detection algorithm to the mask image; and generating a contrast edge image by applying said edge detection algorithm to the contrast image.

8. A method for generating a subtraction image in which the contrast image is well registered with the mask image, comprising the following steps: acquiring a mask image; acquiring a contrast image; generating a mask edge image by applying an edge detection algorithm to the mask image; generating a contrast edge image by applying said edge detection algorithm to the contrast image; using a deterministic sign change (DSC) algorithm to compute, for a) each of a plurality of shifted mask edge images and b) the contrast edge image a like plurality of DSC scores; identifying the highest DSC score so computed; and identifying that shifted mask edge image which corresponds to said highest DSC score.

9. A method for generating a subtraction image in which the contrast image is well registered with the mask image, comprising the following steps: acquiring a mask image; acquiring a contrast image; generating a mask edge image by applying an edge detection algorithm to the mask image; generating a contrast edge image by applying said edge detection algorithm to the contrast image; and using an criterion to identify, for each of a plurality of shifted mask edge images, that shifted mask edge image which is best registered with the contrast edge image.

10. Digital subtraction angiography apparatus for generating a subtraction image in which the contrast image is well registered with the mask image, comprising: means for acquiring a mask image and a contrast image; means for defining, in the contrast image, a region of interest ("ROI") ; means for generating a mask edge image by applying an edge detection algorithm to the mask image and a contrast edge image by applying said edge detection algorithm to the contrast image; means for computing, using a deterministic sign change (DSC) algorithm, for a) each of a plurality of shifted mask edge images each being bounded by an ROI-shaped region and b) that contrast edge image which is bounded by the region of interest a like plurality of DSC scores; means for identifying the highest DSC score so computed; and means for identifying that ROI-shaped region which corresponds to said highest DSC score.

11. The apparatus of claim 10, further comprising means for creating a subtraction image from that portion of the contrast image which is bounded by the ROI and that portion of the mask image which is bounded by said identified ROI- shaped region.

12. The apparatus of claim 11, further comprising display means for displaying said subtraction image.

13. Digital subtraction angiography apparatus for generating a subtraction image in which the contrast image is well registered with the mask image, comprising: means for acquiring a mask image and a contrast image; and means for generating a mask edge image by applying an edge detection algorithm to the mask image and a contrast edge image by applying said edge detection algorithm to the contrast image.