Angiographic Visualization Of The Coronary Arteries In Computed Tomography Angiography With Virtual Contrast Injection
A Löfving, X Tizon, A Persson, Smedby
Keywords
computed tomography angiography, coronary artery, maximum intensity projection, post-processing, visualization
Citation
A Löfving, X Tizon, A Persson, Smedby. Angiographic Visualization Of The Coronary Arteries In Computed Tomography Angiography With Virtual Contrast Injection. The Internet Journal of Radiology. 2005 Volume 4 Number 2.
Abstract
Computed Tomography Angiography (CTA) is a non-invasive alternative to catheter angiography but has had limited use in the coronary arteries due to resolution and visualization problems. Maximum intensity projection (MIP) is difficult to apply in the heart with numerous overlying contrast-filled structures. The study proposes virtual contrast injection, originally introduced for MR angiography, for visualizing the coronaries in CTA. Examination after i.v. contrast injection was made in a 16-slice CT. After virtual contrast injection (concurrent grayscale connectedness) separating the vessels, rendering of the right and left coronary separately was made with MIP and volume rendering (VRT). Images similar to invasive angiography were obtained in all 7 projections used in the clinical routine. The observer also has the possibility to freely select an optimal projection. After validation in larger materials, this may become a clinically useful non-invasive alternative for examining coronary arteries.
Introduction
For some time now, the method of choice for diagnosis of coronary artery disease has been X-ray angiography, including cardiac catheterization. Steadily growing, the expected number of procedures in the USA in 2010 will be 3,000,000, making it one of the most common in-hospital operative procedures [1]. It is to date the most reliable, accurate and reproducible technique available for quantitative evaluation of coronary artery disease. The purpose of the procedure is to define the anatomy of coronary arteries and to measure the degree of luminal obstruction. A great advantage is the possibility to immediately supplement the diagnostic procedure with therapeutic intervention.
The most obvious drawback of catheter angiography is the fact that it is highly invasive and carries a not negligible risk of complications. Estimates of the risk of major complications have yielded figures around 2% [1]. Another limitation is the inability to produce images in new projections once the examination has been terminated. This means that appropriate projections have to be selected during the investigation, usually with a limited number of highly standardized angulations, typically 5 projections for the left coronary arteries and 2 for the right coronary artery. This demands a high degree of expertise of the examiner.
A final drawback is the cost involved, which is in general higher than for all non-invasive procedures.
In recent years, the technical development of multi-slice computed tomography, with substantially improved resolution in all three dimensions, has lead to the introduction of CT angiography (CTA) as an alternative to invasive imaging of the coronaries [2,3,4]. The major advantage of this technique is its less invasive nature (only an intravenous contrast injection is needed), which should result in a lower risk of complications.
Multi-slice acquisition in spiral CT results in a three-dimensional (3D) dataset that has to be presented on a two-dimensional (2D) device such as a monitor. For this presentation (rendering), a number of techniques are available. In multi-planar reformatting (MPR), a section through the volume is constructed and displayed on screen. The Volume Rendering Technique (VRT) allows the user to see a view of the heart from the outside, with the coronaries visible on the outer surface of the myocardium. Neither of these presentation modes, however, resembles the way the coronaries are displayed at invasive angiography. For clinical users, it would probably be an advantage if images could be presented in a projection mode, mimicking that of invasive angiography.
In other vascular territories, Maximum Intensity Projection (MIP) is commonly used with CTA or MR angiography (MRA) data, producing images closely resembling those from invasive angiography. However, this approach is not easily applicable in the coronaries, due to the contrast-filled cardiac chambers and great vessels that will inevitably dominate a MIP image constructed from a region comprising the entire heart.
For analogous problems in MRA, the post-processing technique “virtual contrast injection” was developed [6, 7]. In selected cases, this resulted in efficient elimination of overlying contrast-filled veins. The aim of this study is to ascertain whether the same approach can be applied to 3D CTA data to separate the coronary arteries from other contrast-filled structures and thus permit coronary artery visualization in an angiographic mode similar to invasive X-ray angiography.
Material and Methods
The concept of Grayscale Connectedness generalizes to grayscale images the idea that, in a binary image, two voxels are said to be connected if they belong to the same connected component, i.e. if they can be joined by a path consisting of voxels inside the object [11]. As computed by the grayscale connectedness algorithm, two voxels can have a connectedness value anywhere between zero (not connected) and one (fully connected), depending on the grayscale variations along all possible paths joining them (cf. Fig 1). The grayscale connectedness algorithm uses several sets of voxels, the so-called seed voxels, as starting points for a procedure that has similarities with Region Growing [10]. However, unlike Region Growing, which is a local procedure, grayscale connectedness takes into account the whole path between two voxels to decide if they are connected or not. The spread of gray values from neighbor to neighbor, starting from user-defined locations, has been termed “virtual contrast injection” [6].
Figure 1
For separating coronary arteries from other high-intensity structures such as bones or contrast-filled vascular structures, the grayscale connectedness algorithm is applied in a competitive mode, so that each voxel is assigned to the seed region with which it attains the highest grayscale connectedness value. The result of the procedure is a partitioning of the volume image, where each component consists of one or more vessels (or some other irrelevant high-intensity structure) with parts of the background surrounding it. This is analogous to the problem of separating arteries and veins in MRA images [6, 7].
The processing procedure can be divided into three separate parts.
SeparaSeed Software
The SeparaSeed interface consists mainly of two windows (Fig 2). The left window is a MIP projection with the slices superimposed on each other (collapsed MIP). The right window shows the individual images in the 3D volume, allowing the user to browse through the data set.
Figure 2
A paint tool is available to mark voxels belonging to objects of interest. A separate color is used for identifying each seed region. Color and thickness of the paintbrush can be selected, and the user can paint either in the MIP or in the original 3D volume. Painted pixels in the MIP are relocated to the appropriate slice using a method described in [8]. For visualization of the coronary arteries, one color is used for the left coronary, one for the right coronary and other colors for the atria, ventricles, great vessels and the thoracic wall. In this study, we used red for the left atrium and ventricle, green for the right atrium and ventricle, blue for the right coronary artery, cyan for the left coronary arteries, yellow for the lungs and other structures of no interest, such as the thoracic wall.
Grayscale Connectedness (GC)
This procedure takes, as input, two volumes. The input volume is the original CT stack, and the seeds volume is the color stack produced showing the seed voxels. It is typically sparse, containing starting points for the grayscale connectedness algorithm. The iterative propagation of grayscale values is carried out with a chamfer algorithm [12]. It is terminated when the number of voxels having changed color at the end of each iteration falls below a user-specified threshold. It results in a complete partitioning of the input volume, i.e. each voxel is assigned a color corresponding to one of the seed regions.
GC Visualization
The last step consists in masking the original volume with the results from the grayscale connectedness segmentation. It uses as input the original CTA stack and the output stack from GC. The user selects the number corresponding to the color used for a specific structure, e.g. 1, corresponding to Red, for the left coronary. When this step is performed, all the voxels that, after segmentation, do not match the chosen color are zeroed out in the CT stack. The result is shown as a collapsed MIP, but the output volume can then be presented with more advanced visualization tools, like MIPs in other orientations or Volume Rendering Technique (VRT).
Implementation
The algorithms described in this article have been developed in JAVA as plugins to the widely used Image Analysis program ImageJ [5]. With this solution, it is possible to use the interface on most computer platforms (including PC, Macintosh, Linux) without the need to adapt the code. This plugin is made freely available from the authors.
Material and Data Acquisition
For the preliminary set we used the images from a 69 year old male with a clinical diagnosis of atrial fibrillation. The CT images were collected using a spiral scan with 500mAs/rotation and 0.42s/rotation on a Siemens Sensation 16. We used a primary collimation of 16 x 0.75 mm with a reconstructed slice thickness of 1mm and an increment of 0.5 mm. The matrix resolution used was 512 x 512.
The images are retrospectively EKG triggered, reconstructed in diastole. The triggering time, which is selected individually for each patient, was between 50 and 55% of the time between the R peaks. The temporal resolution, which is EKG dependent, was approximately 200 ms.
Results
An example of resulting MIP images in standard projections is shown in Figure 3A-G. MIP images in different projections, not attainable with invasive angiography, are shown in Fig 4A-B. In Fig 5, the same projections as in Fig 3A-C are presented with VRT. Figure 6 illustrates the ability to select an appropriate projection after the separation has been performed.
Figure 3
Figure 4
Figure 5
Figure 6
Figure 7
Figure 8
Figure 9
Figure 10
Figure 11
Figure 12
Figure 13
Figure 14
In the images shown here it took about 30 min to draw the seeds in the structures. This varies greatly depending on the number of structures to be painted and on the quality of the acquired CTA images. The easier it is to locate the structures, the easier it is for the algorithm to calculate their connectedness. The computer processing time was 10–15 min on a 1.9Ghz P4 with 1024Mb of RAM. This time is also highly variable depending on the number of images, the quality of them and of the matrix size. With a faster computer, the time can be reduced significantly.
Discussion
Judging from recent studies concerning data acquisition with multi-slice CT [2,3,4, 9], it seems likely that CTA will establish itself as an important technique, complementary to invasive catheter X-ray angiography, for imaging the coronary arteries. The use of CTA is believed to reduce the overall risk for the patients at the same time as it provides better insight into the intricate 3D geometry of coronary anatomy. Furthermore it offers much greater possibilities for post processing of the acquired images.
The fact that the most commonly produced coronary CTA images, such as VRT from the outside of the heart, are in a format unfamiliar for cardiologists and thoracic radiologists, emphasizes the need for angiographic projections views. The results shown in Fig 3A-3G show that such a presentation mode is indeed possible with CTA data. In the present study, a single intravenous injection, rather than several intra-coronary injections, was sufficient to produce images for all the projections used in clinical routine at our institution. Whether the image quality in these images will be sufficient for clinical applications is a question that remains to be studied. Figure 3E shows that in this case, the method failed to correctly segment one of the branches of the circumflex artery, giving it a discontinuous appearance. Possibly, this would have been avoided with better quality images.
In addition, the free manipulation of CTA images separated from other high-density structures allows the operator to view the vessels from angles not possible with the standard technique (cf. Figure 4) at no extra risk for the patient but still with the ability to exactly duplicate the standard projections. It also allows the observer to see anatomical anomalies not perceived with standard techniques due to poor visibility. An obvious example would be the possibility to look at structures from a dorsal view, not having to tilt the patient to avoid the spinal column, which can be a problem especially with the LAO (Left Anterior Oblique) projection. The possibility to interactively select the optimal projection angle after the end of the examination seems to represent another important advantage compared to the invasive technique. The time for post processing and evaluation of the post processed CTA images may seem long, but should be compared with the total time consumed in an invasive X-ray angiography session.
The free choice of projection may also allow the operator to make more accurate measurements of specific parts of the coronary anatomy, something that might prove valuable in the future for prognostic purposes. It will also open up the possibility for the image data to be reviewed by additional physicians who can freely manipulate the images as if they where performing the examination themselves rather than relying on a limited number of projections acquired during an invasive X-ray session.
It is also important to point out that the separation technique can be used freely with other post-processing techniques such as VRT. We have chosen MIP since the method of presentation is very similar to the presentation mode that cardiologists and thoracic radiologists are accustomed to. Holding on to MIP presentation allows the examiners to use their experience with little or no effort in learning a new mode of presentation.
Our preliminary implementation of the technique is associated with a number of limitations. The first and most obvious is the quality of the CTA images acquired. The resolution and the presence of artefacts in the original image volumes are of special interest since they seem to greatly affect the quality of the post-processed images. With the use of newer and better CT equipment it will be possible to gather higher resolution images. This will allow the algorithm to separate finer structures. It will also help in producing better MIP and VRT images.
As some of the patients have undergone thoracotomy, they have metal sutures in the sternum creating beam hardening artefacts that may obscure the view and affect the ability of the algorithm to perform a correct segmentation. These artefacts can create a bright link between two structures of interest, tricking the algorithm into finding a false path resulting in higher connectedness between two unconnected structures. The distribution of contrast can also create artefacts since high concentration of contrast medium in the right heart can produce beam hardening artefacts.
File format conversion issues between clinical workstations, rendering workstations, and the ImageJ platform are trivial problems that may hamper the use of the software in its present version. Concerning the design of the Graphical User Interface (GUI) and the implementation itself, a number of feature additions are needed to make the platform better suited for clinical use.
Potential Improvements
The possibility to delete seeds drawn in the Separaseed GUI will help substantially since at present, some small mistakes can only be rectified by starting the seeding all over, consuming a lot of time. A possible enhancement would be to introduce a new kind of seed: a stopping seed. It would not, as with the other seeds, be allowed to propagate. As such, it would impose a boundary between two regions. It could act like a stopping criterion, to limit seed propagation to a defined region of interest. It would also prove valuable when two adjacent structures with similar connectedness values are competing for the propagation area, by controlling the possible leakage of all region growing-like algorithms.
When the contrast is poor between two structures, GC will sometimes fail to discriminate them, and provide erroneous segmentation. It can be understood as a failure from the imaging process, as two different structures appear with neighbouring greyscale values, a typical consequence of Partial Volume Effect. The ability to manually correct the segmentation result is a necessity in this case, as no algorithm is able to perform perfectly.
Depending on the quality of the images, it is sometimes necessary to adjust image parameters such as brightness, which is not possible today. It would also be of value to be able to zoom in the picture where the seeds are to be placed, as some of the structures are very small or are very close to another area, marked with a different seed. Hopefully, this could improve the quality of the separation between structures especially if the input images have a high resolution.
Another area that will need further study and improvement is the design of an intuitive Graphical User Interface (GUI). The GUI used today is crude and needs some work to perform well with staff who are not very accustomed to computers.
In future research, the application of the method to clinical practice will have to be examined. How accurate the technique is compared to invasive X-ray angiography is a question that will have to be scrutinised in detail. It should also be compared to other techniques for improving the visualization of coronary arteries from image stacks, e.g. the “soap-bubble” method [13].
Conclusion
The technique proposed in this study, combining CTA images with the MIP visualization, brings together the advantages of two different situations: separating the coronary arteries from other structures, making them easy to identify, and presenting them in a format familiar for someone used to invasive coronary X-ray angiography. For use in clinical practise, the technique needs to be validated by assessing its agreement with catheter angiography.
Correspondence to
Örjan Smedby Dept. of Radiology / CMIV University Hospital SE-581 85 Linköping Sweden Tel +46-13-22 27 17 Fax +46-13-22 17 99 Email: orjan.smedby@imv.liu.se