Image Postprocessing in Cardiac Computed Tomography

Published on 24/02/2015 by admin

Last modified 22/04/2025

Print this page

This article have been viewed 1704 times

CHAPTER 12 Image Postprocessing in Cardiac Computed Tomography

The role of CT in the evaluation of cardiac disease is no longer the subject of possible applications, but more in the realm of determining positive and negative predictive values as cardiac CT angiography becomes a central part of cardiac imaging. Whether the application is to determine coronary artery stenosis in native vessels, to determine patency of bypass grafts or stents, or to detect anomalous anatomy, CT has been shown in many cases to be the study of choice. In the emergency department, the power of a 100% negative predictive value promises to lead to a paradigm shift in emergency department triage for patients with chest pain.¹^,² Similarly, a gated CT acquisition has become the standard of care to evaluate aortic root pathology, including valvular disease, suspected dissection, Marfan syndrome, and Loeys-Dietz syndrome.³^–⁵ Other applications, such as preoperative assessment and postoperative evaluation in congenital heart disease, are becoming a well-accepted application for CT angiography.⁶^,⁷

Despite more than 400 articles on the subject and a body of literature that continues to grow, few published data have focused on the technical aspects of different processing techniques for the evaluation of the heart and coronary arteries.⁸^–¹³ The “how-to” of data acquisition (16 vs. 64 vs. 128 detector scanners)¹⁴^–¹⁶ is addressed in detail in the literature, as are contrast delivery techniques (test bolus vs. bolus triggering vs. timed injection),¹⁷^–¹⁹ but the analysis of the resultant CT data has been the topic of only a few articles of note. This chapter provides a systematic approach to data analysis, with emphasis on how to use each postprocessing tool to its greatest advantage (Figs. 12-1 through 12-4), to interpret most accurately an isotropic multidetector CT volume of the heart.

FIGURE 12-1 A-F, This set of images shows the relative utilities of each rendering technique. A-D, The proper study protocol has yielded a clear view of the right coronary artery (arrowheads) with MIP imaging, but with MIP it is hard to be certain that one is actually imaging the origin of the right coronary artery. E and F, With VRT, this is not an issue, and the color-coded VRT images provide a better three-dimensional perspective. The VRT images are the same projection, but use different rendering parameters.

FIGURE 12-2 A-D, VRT (A-C) and CPR (D) techniques are used in this case to define an anomalous left coronary artery (arrow) arising off the posterior aspect of the main pulmonary artery. The unusual location was impossible to define on routine axial scans or with routine MPR imaging.

FIGURE 12-3 A-G, In this case, evaluation of an anomalous right coronary artery off the left cusp reveals the value of each technique. A, On a single axial image, one can define the vessel (arrow), but would need a series of individual images to delineate its course. B, With MPR images rendered in the optimal perspective, the origin of the right coronary artery is seen. C, MIP images show the vessel coursing between the aortic root (AO) and pulmonary outflow tract (P), but the full course of the vessel is hard to see without use of very thin slabs. D-G, VRT is ideal for complex vascular anatomy, and color (D and E) or gray-scale mapping (F and G) can be used.

FIGURE 12-4 Schematic representation of the value of CPR in imaging complex vascular anatomy. A, Axial sections show only segments of the artery in cross section. B, Multiplanar reconstruction in the orthogonal plane displays only various oblique cross sections of the artery. C, CPR generated by center-line tracking captures and displays the entire artery.

TECHNICAL REQUIREMENTS

With respect to data analysis, the most comprehensive article has been by Ferencik and colleagues.¹¹ Two important statements from this article are as follows:

“The evaluation of multidetector CT coronary angiography with interactive image display methods, especially interactive oblique MPRs [multiplanar reformats], permit higher diagnostic accuracy than evaluation of prerendered images (curved MPR, curved MIP [maximum intensity projection], or VRT [volume-rendering technique] images).”

“Interactive evaluation of multidetector CT coronary angiography data sets on a workstation should thus be the preferred way of interpretation.”

The key point made by this article was that the method of data set evaluation (interactive interrogation of the volume by the interpreting physician vs. review of preset images generated by an individual not performing the interpretation) is perhaps even more important than the postprocessing tools used. Interacting with the data set using only axial images and MPRs was more efficacious than review of preset three-dimensional images. When preset images are reviewed, the information seen depends on the skill of the individual (interpreting physician or technologist) who performed the postprocessing. We have found that in nearly every case, errors in postprocessing can result in key mistakes in data interpretation. Interactivity means that the interpreting physician must process the images himself or herself. The following rules are suggested to do this successfully:

1 Interpreting physicians should learn how to use a workstation for cardiac-specific applications.

2 The ideal workstation should have dedicated tools for cardiac evaluation, including but not limited to automated vessel segmentation (with minimal user time), stenosis calculation software, vessel tracking software, four-dimensional motion studies, excellent and easy-to-use filming, archiving, and networking solutions.

3 Interpreting physicians should be involved in the full scope of the imaging process, ranging from protocol design to quality assurance.

4 Interpreting physicians must be familiar with the various rendering techniques, including MPR, MIP, and VRT, and must know how to use these techniques, their advantages, and their disadvantages.

5 Another area of practical concern is the quality of the data set to be analyzed. The ability to acquire routinely a high-quality data set is mandatory to doing a quality interpretation. Although scan protocols are beyond the scope of this chapter, they cannot be overemphasized. The proper gating technique (retrospective or perspective), control of the patient’s heart rate, and timing of delivery of contrast material (contrast agent and saline flush) and data acquisition must be optimized.

Let us assume that the patient has been scanned, the data have been processed into the appropriate slice thickness (0.6 to 0.75 mm) and interscan spacing (0.4 to 0.5 mm), and the data have been sent to the workstation. Now what? That question is guided in part by individual preference and by workflow, and partly dictated by the workstation available to the user. Regardless of the workstation specifics, some general principles can be implemented across almost all current platforms.

TECHNIQUE

Axial Images

Although a cardiac CT scan typically is reviewed at least using multiplanar reconstruction and three-dimensional renderings, the axial images (see Fig. 12-3) remain a crucial part of the study workflow. At some sites with highly experienced interpreting physicians, interpretation is typically limited to the axial images with sparing use of additional views in selected cases. On a workstation with a composite display of axial, coronal, and sagittal images combined with a three-dimensional image, our strategy for axial images is to scroll interactively through the data set in what the scanner selects as the optimal phase (0% to 90% of R–R interval on retrospective gated study) for coronary artery visualization. We then do the following:

1 Scroll through the data set to define the quality of the study, including contrast delivery timing, lack of motion artifact, and coronary artery opacification. Based on a quick look, one can quickly ascertain how good the study is technically. We also ensure that the “optimal” phase was selected because occasionally multiple phases need to be processed and viewed.

2 Scroll through the heart in a global sense with analysis of chamber size, and a look for any abnormalities including the pericardium and the paracardiac regions. We also note the size of the ascending and descending aorta.

3 Scroll through the coronary arteries, and look for the origin of each vessel and its branching pattern. Vessels such as the sinoatrial nodal branch and atrioventricular nodal branch are often best seen on the axial images. We follow each coronary artery to look at lumen diameter and determine the presence or absence of calcified or noncalcified plaque. Axial images often nicely define noncalcified plaque especially when lipid-rich. Axial images also define the presence and extent of calcification, although a wide window often is needed in the presence of extensive calcification.

The limitation of axial images is that one may need to look at 200 to 350 individual slices to get through a single vessel, such as the left anterior descending coronary artery. Vessels with off-axis courses, such as the right coronary artery, are best defined when one is looking beyond the axial plane. A more global presentation of an entire coronary artery is impossible without a curved planar reconstruction (CPR) (see Fig. 12-4) or three-dimensional display.

Our experience is that referring physicians desire three-dimensional maps to correlate with what they typically see on classic catheter angiography. Axial CT images in this regard are not helpful and do not provide similar information.

Although we typically scroll through the classic axial display interactively, other physicians believe that reading studies can be noninteractive, and that there are no data to prove otherwise. This controversy needs to be addressed in a future study. One issue is the lack of a well-designed, controlled study on the use of different techniques for coronary CT angiography analysis. This is in contrast to many articles published on reading techniques (e.g., two-dimensional vs. three-dimensional) for virtual colonoscopy.

Although most physicians review the axial images as slabs in the plane acquired, one helpful hint is to draw a line through the plane of the aortic valve, and scroll up and down through this plane. This practice often provides a perfect en face display of the proximal coronary arteries.

Multiplanar Reconstruction—Coronal and Sagittal Images

The use of coronal and sagittal reconstructions is common in a range of CT applications, and in cardiac CT its role is also defined. With isotropic data, the individual cardiac chambers, aorta, and pulmonary vasculature can benefit from these displays, and they are routinely used. Whether it is for suspected aortic dissection or suspected pulmonary embolism, these displays are essential. On most workstations, the thickness of the reconstruction can be interactively adjusted (Fig. 12-5), but when slabs are reduced to beyond a few millimeters, the images become fuzzy and are of no value. This is in contrast to MIP images, where often less data may prove to be useful for these large-volume applications. One issue with coronal and sagittal reconstruction and evaluation of the coronary arteries is that because the vessels have a nonlinear pathway, only short segments can be analyzed at a time. When the left anterior descending coronary artery is analyzed with a coronal or axial display, one must constantly change the angle of the reconstruction plane to follow the vessel. Changing the angle can be done interactively, but potentially results in more room for error and less than ideal image displays. For analyzing selected segments of a vessel, this technique might be perfect. If there is a suspect area on the axial views, the coronal or sagittal view may help clarify and quantify this potential narrowing. Key points include the following:

1 Coronal and sagittal imaging may be useful, but it depends on the vessel tracking capabilities of the workstation. Ideally, a specific point is selected, and viewing is adjusted from that perspective.

2 Interactivity is key.

3 The oblique planes are valuable for additional viewing of selected suspicious areas, which often arise after review of the axial images.

4 With coronal, sagittal, and oblique images, one needs to be cautious in grading stenosis and not be fooled by partial averaging. We do not attempt to grade stenosis based only on these images because of this pitfall.

5 We have found that these reconstructions are best for larger vessels, as discussed previously.

6 Many sites now do preprocessed batched coronal and sagittal images at 1- to 3-mm intervals. This may mean a data set of 30 to 120 images that can be scrolled through on a picture archiving and communication system (pacs). Although this technique may be sufficient for routine abdominal or chest CT interpretation, interpreting physicians must recognize the limited utility of this technique for evaluation of the coronary arteries because of the vessel size (typically 3 to 5 mm) and ectatic course of the coronary vessels.

FIGURE 12-5 A-H, Sequential illustrations show different requirements for slab selection when using VRT versus MIP, to visualize vessels distinctly within a volume. A, When the entire volume is volume rendered, vessels are clearly defined and separated because the three-dimensional relationships are maintained. B, With MIP, the vessels are displayed without separation or true spatial orientation. C-E, Reduction of the large volume to a 15-mm slab (C) removes one vessel and results in continued good spatial discrimination with VRT (D). With MIP (E) the vessels remain unseparated without spatial orientation. F-H, Further narrowing of the slab to 5 mm (F) enables VRT (G) and MIP (H) to show the vessels discretely; MIP remains a flat projection that does not convey three-dimensional relationships.

(From Fishman EK, Ney DR, Heath DG, et al. Volume rendering versus maximum intensity projection in CT angiography: what works best, when, and why. RadioGraphics 2006; 26:905-922.)

Curved Planar Reconstruction

As discussed, the limitations with coronal and sagittal reconstruction largely involve the inability to track the complex courses of the coronary arteries. CPR addresses the issue directly (Fig. 12-6). CPR is best thought of as a reconstruction plane that perfectly tracks a vessel regardless of the complexity of its course or direction. In the initial versions of CPR, the user would drop multiple points through the path of a vessel, and the computer algorithm would track the points and create a vessel map. Currently, most vendors have designed software that makes the process more robust and easier to use, and requires minimal user interaction.¹⁰^,¹² On some workstations, there is a program that can be used for tracking the coronary vessels. One simply picks a start and an end point, and the computer automatically tracks the vessel. The vessel is laid out like a string or piece of spaghetti, and the user can rotate the plane to look at the vessel and analyze it from multiple perspectives. The vessel path is calculated as a central line through the vessel, which is also crucial in avoiding errors in analysis. The curved planar images can be viewed in classic soft tissue window or altered to be presented in MIP mode. In cases where there is a gap or stenosis in the vessel, dropping additional seed points can be done to help with vessel tracking.

FIGURE 12-6 A-H, Extensive calcification (arrowheads), as shown in this case, makes it difficult to determine the degree of stenosis. After analysis using multiple postprocessing tools, we are able to define correctly a greater than 70% stenosis in the left anterior descending artery, which was stented. In these cases, we often find a combination of axial (A), MPR (B), and CPR (C-E) images the most helpful. MIP (F) and VRT (G and H) added little to the final diagnosis, but did give a nice representation of the extent of calcification.

We use CPR for every coronary artery CT angiography (Figs. 12-7 and 12-8). The ability to lay out a vessel without the worry of partial averaging is important not only when trying to define the presence of stenosis, but also for calculating the degree of stenosis. On some systems, the CPR also serves as the basis for the system to calculate automatically the percent stenosis present. A function that is common on all cardiac-specific software packages and many vessel analysis packages has the user pick the area of concern (usually area with greatest stenosis) and select a point of normal vessel proximal and distal to the stenosis. When selected, the system automatically calculates the percent stenosis. It is important to be careful with these programs because there are many potential sources of error, but at least they are a guide, and in the future they should be more robust. Several key points regarding CPR and some highlights and pitfalls are as follows:

1 The success of CPR depends on the delivery of a well-timed contrast medium bolus with homogeneous enhancement of the vessel. Values of 250 HU and greater are ideal. With poor opacification, automated vessel extraction programs have issues and result in errors of omission and commission.

2 The success of CPR depends on a good center-line definition. In larger vessels such as the aorta, this is fairly simple. In the coronary arteries measuring 2 to 5 mm, if the center-line trace is inaccurate, it could result in the overestimation of stenosis. A center-line error is most likely to occur in the presence of extensive coronary artery calcification.

3 Although dense calcification can potentially throw off the center-line measurement, CPR is especially valuable in defining the degree of stenosis in these cases. With dense calcification, rotation of the vessel 360 degrees helps with determining whether the calcification is mainly eccentric and representing positive remodeling or causing a critical stenosis.

4 Some newer software versions allow the computer to segment the vessels automatically without user interaction. This may work well in selected cases, but careful attention to recognize incorrect tracking is crucial. When a CPR is viewed in MIP mode, it is advisable to make the vessel thicker when looking at the vessel for potential stenotic areas. If reviewed in standard soft tissue mode, this would not be helpful.

5 CPR often does not work if the patient has an arrhythmia that may result in step-off artifacts in the data set. Similar problems can occur with breathing-related artifacts.

FIGURE 12-7 A-L, Small calcified and noncalcified plaque (purple line in G and H), which do not cause any significant narrowing of the left anterior descending artery, are identified in this case. To supplement the axial images (A and B), MPRs (C and D) and CPRs (E-H) are ideal for image analysis. The small plaque in the left anterior descending artery is seen on the MIP images (I and J), but is hard to define prospectively on the VRT images (K and L). This case also reveals the importance of thin slabs for vessel detection on the MIP images.

FIGURE 12-8 A-K, A patient with a high calcium score (Agatston >500) but without a critical stenosis. A and B, CPR images are best for defining the vessel lumen and the positive plaque remodeling. C-I, A series of MIP images with varying slab thickness of 1 to 20 mm and varying orientations show how difficult it is to use MIP images in the face of calcification (arrowheads). Depending on the view and parameters, the left anterior descending artery, circumflex artery, and ramus intermedius can look occluded or patent. J and K, VRT images help show the eccentric nature of the plaque burden.

Maximum Intensity Projection

The role of three-dimensional imaging, using MIP or VRT, still is controversial today. Although most sites use some type of three-dimensional imaging for selected cardiac applications, experts have yet to reach consensus as to whether this practice is mandatory or necessary in all cases. Our experience is just the opposite, however. Postprocessing with three-dimensional imaging is an essential part of every examination, providing crucial information not revealed by other interpretative methods. One caveat is that MIP is most valuable and accurate in a patient without coronary artery calcification. In the presence of small amounts of calcification, MIP images can easily overestimate the presence and extent of stenosis.

In practice, MIP is usually performed with a sliding slab to look at the coronary arteries. The term sliding MIP is used to emphasize that it is done interactively at the workstation by the interpreting physician to define the best plane for vessel display. There are numerous tricks of the trade and pitfalls with MIP in general.²⁰^,²² By adjusting the MIP slab thickness (1 to 30 mm), one can interactively vary the size of the segmented volume as needed (see Fig. 12-5), but one always needs to be careful not to edit out crucial parts of the data set. The potential issues with MIP can be seen through a careful understanding of the technique itself. Some key points from a technical perspective are as follows:

1 The displayed pixel intensity represents only the material with the highest intensity (i.e., density in the case of CT) along the projected ray. A high-intensity material, such as calcification, would obscure information from an intravascular contrast agent. This limitation can be partially overcome with use of nonlinear transfer functions or, more practically, through volume editing. Volume editing can take the form of a preprocessing step (e.g., section-by-section or “slab” editing) or an interactive process (e.g., “sliding-slab” MIP).

2 Use of the highest intensity of the image in effect also selects the “noisiest” background voxels, and decreases the visibility of vessels in brightly enhancing structures, such as the cardiac chambers and pulmonary arteries.

3 MIP images cannot be displayed with surface shading or other depth cues, which can make assessment of three-dimensional relationships difficult. MIP images can be misleading, and incorrect interpretation can be a result of viewing only the MIP images. Also, volume averaging (the effect of finite volume resolution) coupled with the MIP algorithm commonly leads to MIP artifacts. A normal small vessel passing obliquely through a volume may have a “string of beads” appearance because it is only partially represented by voxels along its length. The severity of these artifacts depends on the resampling filters used.

4 MIP is probably the easiest technique to master because of its simplicity (it uses window width and level), but in practice has a reasonable robustness for many applications. In our experience, MIP is very useful as an adjunct to VRT, and we routinely use MIP viewing in practice. Because of its limitations, however, we would never use MIP alone without evaluation of axial, MPR, CPR, and VRT images.

In a patient without calcification, MIP can create agreeable images of long segments of the coronary arteries that superficially appear as a comprehensive display. Soft plaque is overlooked, however, unless it creates significant lumen narrowing. Rybicki and coworkers ¹³ also noticed this issue and issued the following warning:

“In summary, while thick MIP images are visually appealing and can be used for rapid interpretation algorithms in the emergency setting, they must be used prudently. When the MIP thickness is increased, multiple structures are projected onto the same plane. Distinguishing details between these structures can be lost when the thickness is too large, and, thus, the user must have full understanding of the anatomic detail before MIP images can be safely rendered.”

We use MIP in every case because it does add value, especially in terms of speed and for viewing small-caliber distal portions of smaller vessels. Some key points are as follows:

1 When using MIP, the interpreting physician must be careful not to cut away crucial pathology or create a “pseudolesion.”

2 MIP works best when there is little or no calcification present. In the presence of calcification, MIP can lead to overestimation of stenosis or misinterpretation of an occlusion.

3 One must adjust the slab thickness carefully to see the vessel in full view, but one must be careful not to overlook lesions, especially very proximal lesions.

4 One must never use MIP as the only rendering technique for interpretation of a case. Even when the study looks normal, another display technique should be used for confirmation.

5 MIP works best when used interactively by the interpreting physician. MIP images generated by technologists or physicians other than the interpreting physician may have several of the pitfalls noted earlier.

6 Because MIP relies on the brightest pixels, proper synchronization of contrast medium injection and data acquisition is crucial to maximize image contrast resolution.

7 In selected cases, one can go directly from the axial images to MIP to generate a vascular map, especially when the axial images suggest a normal study without any plaque burden.

Volume Rendering Technique

VRT has a crucial role in evaluation of cardiac and coronary CT angiography (Fig. 12-9). The role of VRT in the analysis of these cases is based on numerous factors, including the ability of VRT to display large volumes of data accurately, especially from complex data sets. The best way to understand the crucial role of VRT is to remember several of the basics of VRT.²⁰