Applied Radiology

Dr. Hsu is currently Chief Resident in Diagnostic Radiology at Johns Hopkins Hospital. He completed internship at the University of Hawaii, and graduated from Washington University School of Medicine in St. Louis, where he was a Howard Hughes Fellow. Dr. Hsu will begin a Body Imaging fellowship at Stanford in 2003.

Image quality in computed tomographic angiography (CTA) is critically dependent on contrast injection, data acquisition, and postprocessing technique. First, injection parameters are chosen to achieve intense homogeneous enhancement despite inter-individual hemodynamic variability. Next, CT data acquisition must obtain adequately thin sections over the appropriate volume of interest during a limited temporal scanning window. Finally, reconstruction and rendering techniques must be rationally selected to demonstrate relevant anatomy and disease processes in an accurate manner. Advances in these areas and in scanner technology will enable the development of new applications. This article provides a primer for conceptual understanding of parameters that allow optimization of CTA studies.

Computed tomographic angiography (CTA) is a minimally invasive technique used to study intracranial, thoracic, abdominal, and peripheral vessels. The utility of a number of applications has equaled or surpassed catheter angiography. In comparison with catheter angiography, CTA offers more viewing angles with simultaneous evaluation of vascular lumen, wall, and surrounding soft tissue, yet is less expensive and time consuming, with fewer complications and usually with lower radiation dose. In comparison with magnetic resonance angiography (MRA), CTA usually offers better resolution and less artifact, although ionizing radiation and iodinated contrast are required.

Image quality is critically dependent on contrast injection, data acquisition, and postprocessing technique. In each of these areas, conceptual understanding of the effects of various parameters is necessary for optimization of CTA studies. In addition, advances in these areas are facilitating improved diagnostic quality and the development of novel applications.

Contrast injection

Image quality starts with the mechanical injection of iodinated contrast. Injection techniques have been developed to achieve optimal enhancement and increase the efficiency of contrast utilization. This, in combination with dramatically faster scanners, allows the use of much lower contrast doses than just a decade ago.

Concept-- The ideal vascular enhancement pattern is homogeneous and intense, delivered to match the timing of data acquisition.

The goal of contrast injection is to achieve a relatively homogenous plateau in intravascular enhancement throughout the duration of data acquisition. This focus on prolonged homogenous plateau differs somewhat from conventional angiography, where less inherent image contrast makes a high narrow enhancement peak preferable to maximize vascular opacification. ¹

Plateau homogeneity can be characterized by "plateau deviation," defined as the standard deviation of the time-attenuation curve in an intravascular region of interest (ROI) during the scanning period. ² Homogeneous enhancement improves diagnostic quality of images and facilitates image processing. Non-uniform enhancement can result in artifactual pathology such as filling defects and perceived stenoses. Inhomogeneity also complicates image processing, particularly parameter selection for three-dimensional (3D) rendering. ³

The interplay of multiple injection and acquisition parameters must be understood to overcome intra- and inter-individual variability in vascular time-attenuation curve. Achievement of a homogeneous temporal scanning window improves image quality, while efficient utilization of injected contrast decreases required dose.

Concept-- Independent injection par-ameters alter intravascular bolus geometry.

The effect of various injection and patient parameters on the ultimate intravascular geometry of a monophasic contrast bolus was presented by Cademartiri ⁴ at the annual meeting of the Radiological Society of North America (RSNA 2001). Time-enhancement curves were constructed for monophasic injections, examining the effect of injection volume, rate, and concentration; use of saline flush; and patient size and cardiac output. Peak enhancement (PE) and time to peak enhancement (tPE) were specifically examined, with some consideration of plateau characteristics.

Volume: Increasing volume of injection while maintaining constant rate and iodine concentration results in increased PE and lengthened tPE, primarily via summation and potentially by recirculation. The intravascular curve becomes higher and broader with a delayed peak. Of course, the maximum injected volume is limited by iodine load. A practical application of the delayed peak phenomenon is necessary when a test injection is used to plan scan timing. Time to peak enhancement measured with the small test injection will underestimate tPE of the actual CTA injection, and scan timing should be appropriately delayed. ⁵

One simplified method for calculating the appropriate delay is to assume a fixed "circulation time" from the center of the bolus to the time of PE. For the high volume injection, the "circulation time," calculated from the test injection, is added to the temporal center of the bolus to predict the time of PE. Scanning is then centered on the predicted peak.

Rate: Increasing only the rate of injection with constant volume and iodine concentration results in increased PE, decreased tPE, and shortened plateau. For practical purposes, this means that scanning must be accomplished earlier, more quickly, and with more precise timing. One application of these effects is when CTA acquisition is performed with fixed postinjection delay rather than tailored delay. In this case, a less rapid injection can be utilized to achieve a broader, and therefore more forgiving, time-attenuation curve.

Iodine concentration: Increasing iodine concentration with constant rate and volume results in increased PE with no change in tPE. This concept may be employed to increase arterial attenuation without altering the temporal course of enhancement, subject to limits on iodine load and FDA-approved concentrations (up to 370 mgI/mL).

Concept-- Immediate saline flush, though not routinely employed, uses contrast to better advantage.

Following contrast injection with immediate saline flush uses injected contrast more efficiently by significantly decreasing venous pooling in the injected extremity. This results in increased PE, delayed peak, and prolonged plateau. Use of saline flush can thus improve the vascular-enhancement profile with a fixed contrast dose, and this technique can, in fact, be used to decrease contrast dose. ⁶ Use of flush has the additional advantage of decreasing streak artifact, particularly in thoracic CTA, where there can otherwise be high residual contrast concentration in the subclavian vein ipsilateral to an antecubital injection. Of course, excess flush will also dilute the contrast bolus.

Commonly, approximately 10 mL isotonic saline is injected manually, although the timing and rate of a manual injection is neither ideal nor reproducible. Alternatively, attempts to layer saline and contrast in the same syringe have proved tedious and unreliable. The best way to achieve a rapid and reproducible flush would be to use a double-barreled power injector. One experimental protocol using 30 mL flush with such an injector demonsttrated a 20% decrease in contrast material requirement. ⁶ Unfortunately, double-barreled injectors used for MRA are of insufficient volume for CTA, and to our knowledge, a system adequate for CTA is not currently commercially available in the United States.

Concept-- Inter-individual differences in intravascular time-attenuation re-sponse presents further challenges to
the selection of optimal injection and acquisition parameters.

Patient size: Greater dilution of contrast will occur in a patient of greater size and circulating volume, potentially necessitating a higher contrast dose to maintain image quality. ⁴

Cardiac function and age: In a patient with high cardiac output, the bolus is both circulated and diluted more rapidly, resulting in a narrower time-attenuation curve. With decreasing cardiac output and increasing age, the bolus is circulated less rapidly and with less initial dilution. Consequently, scan timing and contrast dosage can be rationally adjusted.

The ratio of cardiac output to weight has been used to predict the kinetics of arterial enhancement. ⁷ However, noninvasive estimation of cardiac output can be quite unreliable.

Concept-- Tracer kinetic theory can be applied to analyze injected bolus geometry, vascular response function, and resultant intravascular time-attenuation curve.

An "impulse" of tracer is a theoretical infinitesimal test bolus at the point of injection. The vascular "impulse-response function" is the resultant intravascular time-attenuation curve in a particular vessel after an injected impulse. The impulse-response function of a particular vessel is determined by mathematical deconvolution, after measuring an actual test injection and the resultant vascular response.

Knowledge of the impulse-response function of a vessel is useful in the following manner: At the point of injection, the actual injected bolus has known geometry that can be represented by a temporal summation of impulses. Therefore, the actual resultant intravascular time-attenuation curve can be predicted by a similar summation of impulse-response functions. Conversely, bolus geometry may be designed to achieve a desired intravascular time-attenuation curve.

Fleischmann provides a more detailed technical discussion of tracer kinetics relevant to CTA. ⁸

Concept-- With knowledge of patient impulse-response function, injected input function can be tailored for optimal intravascular bolus geometry.

Several strategies can be employed in tailoring injected input function to optimize time-attenuation curve in the vessels of interest. Although full application of tracer kinetic analysis is not common clinical practice, limited use of these concepts is valuable for altering injection protocols.

Monophasic injection: This refers to a constant injection rate for the duration of the injection. Of course, injection rate, duration, and concentration can be varied between injections, as discussed above. For practical reasons, this technique is most commonly employed, with an injection rate of 3 to 4 mL/sec. However, with constant injection rate, the arterial time-attenuation response progressively increases and reaches peak just after completion of the injection, followed by relatively rapid decrease in intravascular attenuation. ^5,9 In other words, monophasic injections result in a non-uniform vascular enhancement "hump" during image acquisition, without a true plateau phase.

Biphasic injection: This technique was developed to achieve more favorable plateau characteristics. While monophasic injection results in a "hump" profile, well-designed biphasic injections generally result in a gentle "saddle" profile. Usual biphasic schemes utilize an initial rapid phase followed by a slower phase of injection, resulting in a less dramatic peak, often with a second peak and a more prolonged and homogeneous plateau, which includes the gentle inter-peak valley. Fleischmann ² reported the use of biphasic injections, individually tailored based on Fourier analysis of a test injection. This approach resulted in significantly decreased standard deviation of attenuation during scanning and moderately decreased inter-individual variability. Most commercially available CTA power injectors are capable of biphasic injections, programmable with standard software.

Complex/multiphasic injection: This technique further tailors the injected input function to achieve a more prolonged and homogeneous plateau. Bae ⁹ designed a physiologically based pharmacokinetic model that predicts that a multiphasic injection with exponentially decreasing rate provides the most uniform vascular enhancement. Optimal decay coefficients were found to be proportional to cardiac output per body weight. Using proprietary software and injectors, uniform prolonged enhancement was demonstrated using an animal model, ⁹ and similar results in human subjects were presented at RSNA 2001 with quantitatively better time-attenuation curves than monophasic and biphasic injections. ¹⁰ In practice, unreliable estimation of cardiac output may complicate selection of decay coefficient, therefore, the addition of a test injection could be beneficial.

Use of properly selected biphasic and multiphasic injections ultimately results in more uniform and prolonged enhancement plateau and more efficient use of iodinated contrast, with the potential for improved image quality and decreased contrast dose requirement. With technical improvements in power injectors allowing more precisely controlled and predictable injection profiles, the benefits of tracer kinetic analysis will become more fully realized, particularly as more demanding applications evolve. In practice, further studies are needed to demonstrate clinically significant improvement in image quality for specific CTA applications and to streamline the process of bolus tailoring.

CT data acquisition

The development of spiral CT, resulting in volumetric acquisition and increased speed, was a prerequisite for CT angiography. Continued advances over the past decade include faster gantry rotation, complex interpolation algorithms, and X-ray tube design with high heat capacity and high photon output, enabling both prolonged and faster scanning. The recent introduction of multidetector-row CT (MDCT) has further increased scanning speed, allowing shorter acquisition times, greater volume coverage, finer z-axis resolution, and decreased contrast requirement.

For specific applications, various interrelated factors must be considered in optimizing data acquisition. These include timing, scanning range, collimation, and pitch.

Concept-- Scanning should be performed during intravascular enhancement plateau, but before significant venous and tissue enhancement.

For arterial CTA, scanning should be performed during the first-pass arterial phase enhancement plateau, before venous contamination and significant tissue enhancement. Wide inter-individual variation can make this a challenging proposition, particularly for highly vascular organs where parenchymal and venous enhancement occurs relatively early. In one series, the mean time between an aortic enhancement of 50 HU and the same enhancement in the pancreas was <14 seconds. ¹¹ Strategies for timing the commencement of scanning include fixed post-injection delay, preliminary test injection, and real-time radiographic bolus tracking.

Fixed delay: Scanning after a fixed post-injection delay is the most convenient and widely used timing method, with adequate results in most cases. The delay may be tailored based on anatomic ROI, desired vascular phase, and patient factors. For example, at our institution, arterial CTA of the chest and abdomen is performed after a delay of 25 to 30 seconds, with a delay of 60 to 70 seconds used for venous imaging, and an additional 10 seconds or more added for older patients and those with decreased cardiac output.

A slightly slower injection may also be employed to achieve a longer but lower plateau, which is more forgiving of hemodynamic variability. Older patients and those with a lower cardiac output also tend to have a longer intravascular enhancement plateau, although there is inherently more inter-individual variability in this population.

Test injection: A preliminary small-volume test injection of 10 to 15 mL can be administered to examine vascular response. The test injection is administered at the same rate as the anticipated injection, with sequential scanning at a single axial level to determine the time-attenuation response in the vessel of interest. As previously discussed, the scanning delay for the actual injection can then be planned. The delay must factor in the higher and more delayed peak and prolonged plateau expected based on the larger volume and duration of subsequent injection.

As detailed above, Fourier analysis of vascular response to a test injection can also be used to tailor injected bolus geometry.

Bolus tracking: Rather than using a test injection, the actual large volume injection can be administered and monitored in real time, and scanning triggered "on the fly" when intravascular attenuation at a sentinel axial level reaches a pre-determined threshold. A relatively low threshold, such as 50 HU units above baseline, is usually chosen to factor in the several second delay before the CTA acquisition can commence. A higher threshold may be chosen for patients with lower cardiac output, who are expected to have a less rapid upslope and delayed plateau. A safety mechanism should be in place to trigger scanning if bolus tracking fails, perhaps due to patient motion or failure to reach threshold.

The test-injection and bolus-tracking techniques do involve additional radiation. These preliminary images can be obtained with reduced tube current, thus contributing a very small percentage of patient radiation dose. These techniques allow more accurate and precise timing, and therefore require a smaller "scanning window." ^11-13 Clinically relevant improvements in image quality for specific applications remain to be demonstrated.

Concept-- Limited scanning window requires efficient use.

The scanning window refers to the time available for scanning. This can be limited by duration of enhancement plateau, hemodynamic considerations, and breath holding. Time required to achieve a particular scanning protocol depends on the desired volume of coverage, as well as collimation, pitch, and scanner properties.

Limiting the scanning range to the volume of interest decreases scanning time, allowing thinner collimation, less motion, and more homogeneity in enhancement. Conversely, requisite thinner collimation for some applications, such as CTA of the circle of Willis, may limit volume coverage. Prolonged scanning time can result in image degradation due to finite intravascular contrast plateau, appearance of tissue and venous enhancement, and respiratory motion.

The development of 4- and 8-row MDCT in combination with faster tube rotation has dramatically increased scanning speed, allowing greater volume coverage and thinner collimation. For example, MDCT has enabled complete lower extremity inflow and runoff studies with a single injection, as well as thin-section CTA covering the entirety of the carotid arteries and circle of Willis. ¹⁴ Sixteen-row MDCT may increase scanning speed even further and/or allow thinner sections, potentially improving image quality by the mechanisms discussed above, and facilitating the development of novel applications, such as coronary CTA.

Concept-- Thin collimation improves resolution but increases noise and scanning time.

Collimation: Collimation of the X-ray beam is related directly to slice thickness, although effective slice thickness will be slightly wider, depending on specific parameters of helical acquisition and interpolation. ¹⁵ Maximizing Z-axis resolution would appear to be an important objective, to decrease partial volume averaging and reduce 3D anisotropy, an important source of artifact in 3D rendering. However, thinner collimation requires increased scanning time to cover a given Z-axis range. In addition, thin collimation increases quantum noise, which may degrade image quality, particularly in obese patients.

In practice, collimation is chosen for each application based on Z-axis resolution requirement, volume of interest, and temporal scanning window. Thin collimation is most important when evaluating small vessels and vessels with a significant oblique or in-plane component, such as renal arteries and the circle of Willis. Tailored use of thicker collimation may be chosen for obese patients, and those with decreased breath-holding capability.

Concept-- Higher pitch improves scanning speed more than it degrades Z-axis resolution.

Pitch: Pitch is defined as table feed per rotation divided by collimation. Increasing pitch means that the table will move further during a single rotation of the X-ray tube, therefore the slice-sensitivity profile is widened, and effective section thickness is increased. However, reconstruction using of 180š rather than 360š interpolation limits Z-axis broadening, enabling good image quality with a pitch as high as 2 with single-detector-row CT (SDCT) scanners. This obviously reduces scanning time, but also increases effective section thickness, albeit to a lesser degree. ^15,16 When pitch is doubled from 1 to 2, scanning speed is doubled, but effective section thickness only increases by a factor of 1.3. It is therefore advantageous to use higher pitch with narrower collimation, to achieve similarly thin sections and still realize a gain in speed and scanning range. Pitch can be increased even further with MDCT because interpolation can be performed across helical channels. However, as pitch increases, peripheral undersampling artifacts occur, and for SDCT a pitch >2 is rarely used. By the same token, for MDCT pitch is usually limited to (2 * N), where N is the number of detector rows.

In the special case of MDCT with adaptive array detectors and adaptive interpolation algorithm, within limits, effective slice thickness does not significantly change with increasing pitch. ^17,18

Brief overview of image reconstruction and rendering

Advanced rendering techniques, particularly those that offer a 3D perspective, facilitate efficient visualization and analysis of an inherently volumetric dataset. These generated images are also helpful tools in communication with clinicians, as they can be made to emulate familiar perspectives, such as catheter angiograms or surgical approaches. In one author's vision, 3D images are "oriented to anatomy rather than the CT table." ¹⁴

Image reconstruction: Image reconstruction using 180š interpolation decreases effective section thickness, thereby increasing Z-axis resolution. By decreasing peripheral undersampling artifacts, it also enables the use of higher pitch, with all of the benefits discussed above, as well as lower radiation dose. ¹⁹

Reconstruction interval can be arbitrarily varied in spiral CT. An overlap of approximately 50% is adequate for most CTA applications. Overlapping reconstruction improves Z-axis resolution ²⁰ and is most important for imaging small vessels and those oriented in the axial plane, such as renal arteries. Due to anisotropic data, CTA is more accurate in assessing luminal diameter of vessels running perpendicular to the axial plane, such as the aorta or carotid arteries. ²¹ Artifacts more commonly arise in CTA of vessels not perpendicular to the axial plane.

The primary disadvantage of highly overlapping reconstruction interval is the time-consuming generation of larger datasets, which may tax capacities for storage and manipulation for 3D reconstructions. Alternatively, selected portions of a study may be reconstructed with more overlap. For this purpose, raw data should be stored until initial images are reviewed and desired modified reconstructions performed.

Source images: Initial CT reconstructions can be performed only in the axial plane. These axial images are traditionally the primary diagnostic tool in clinical CT. These images also serve as the source data for a variety of additional reconstruction and rendering techniques. These techniques each have specific advantages in various imaging scenarios, and should be used in a complementary fashion.

Multiplanar reconstruction: Images can be reconstructed easily in coronal and sagittal planes on most scanners and workstations. These additional orthogonal planes can be used to more easily define and depict the longitudinal extent of a structure or process. Sagittal and coronal reconstructions are particularly useful for vessels running in the axial plane, such as renal arteries. Of course, as long as the source data are anisotropic, Z-axis resolution will remain inferior to X- and Y-axis resolution in the reformatted planes.

Curved planar reformations: This is an additional planar technique in which a curved plane is defined through the center of a curved structure, such as a tortuous vessel. For example, to examine the aorta in a curved coronal plane, the vessel is first viewed in the sagittal plane and a curved plane is manually defined by tracing a curve through the center of the vessel. A curved coronal plane is then generated along the curved line. This process can be performed in any plane. Once the curved plane is generated, some software packages also allow scrolling in that plane, as well as adjustment of slice thickness.

This technique is commonly used in CTA of the aorta, iliac arteries, renal arteries, and carotid arteries, and will be indispensable in coronary CTA. A current limitation of this technique is its dependence on manually placed center points to trace the curved plane. Misaligned points often lead to simulated vascular stenoses in the curved planar reformation. Several manufacturers will soon market automated systems for vessel tracking.

Shaded surface display (SSD): The first 3D rendering technique to be used clinically, SSD employs a HU level threshold or range to depict the surface of an object. A 3D effect is accomplished by displaying the generated surface with shading to emulate the effect of a virtual light source. If the threshold level is well chosen, the surface of the structure can be well defined, with all irrelevant structures excluded. SSD can be constructed with high spatial resolution, and is also the 3D technique that generally presents the clearest depth cues. Because of the inherent data compression, SSD images require relatively little processing power to generate and manipulate.

However, optimal threshold level may be difficult to select, and it may not be possible to simultaneously optimize all structures or segments of interest on a single image. Outside of vascular applications, some structures may simply not have well-defined surfaces. In addition, SSD discards a great deal of data, effectively reducing each voxel to 1 bit of data, ie, whether it is above or below the threshold level. For example, vascular calcifications and enhanced intravascular blood will both be above threshold for SSD, thus leading to underestimation of calcified stenoses.

In comparison to maximum intensity projection (discussed below), SSD better depicts complex branching patterns and spatial relationships. However, many of the advantages of SSD are also captured by volume rendering (see below), a technique which preserves the dynamic range of the source data, and has replaced SSD for most CTA applications.

Maximum intensity projection (MIP): This widely available technique projects a volume of interest onto a specified plane, mapping the maximum intensity in the volume along each ray perpendicular to that plane. Maximum intensity projection was originally designed for CT and MRA, and continues to be used extensively for this purpose.

Inherent to CTA MIP algorithms is the need for data editing, as overlapping dense structures such as bone can exceed the HU values in enhanced vessels. Similarly, because the highest intensity voxels are chosen along each ray, background mean intensity is raised, and noise can also be accentuated. Use of thin volumes of interest helps to reduce reconstruction artifacts. A single MIP image conveys no depth information, and MIP is particularly limited in the definition of overlapping structures. The angle of the projected plane must be rotated to define complex spatial relationships. Use of a cine loop is helpful for this purpose.

Although the concept of MIP bears a slight semblance to SSD in that it is, in essence, a threshold technique, MIP does preserve the attenuation value of the displayed voxel. Therefore, with appropriate windowing, some differentiation between elements of the displayed structure is possible, such as between mural calcification and enhancing blood. Despite its limitations, MIP is generally considered superior to SSD for CTA.

Volume rendering (VR): Unlike threshold techniques such as SSD and MIP, volume rendering preserves all of the data from the source images, and real-time interactive VR has become less cumbersome and more widely available as computer processing power has increased. The concept of VR is to preserve the 3D volumetric dataset acquired via spiral CT and display it as a true volume of data rather than as arbitrary two-dimensional slices.

This is accomplished by selecting a volume to be analyzed, then interactively using various display algorithms to select and weight voxels to achieve a display that highlights tissues and relationships of interest. Transfer functions are used to map properties such as opacity, brightness, color, and windowing to the voxels in the volume of interest, with all voxels in the volume potentially contributing to the final image. In real time, the displayed image can be cut and rotated, and transfer functions can be altered. Calhoun ²² presents a historical perspective of volume rendering and technical discussion of transfer functions.

Real-time interaction with a volume of data offers many potential advantages over threshold techniques, including better delineation of 3D relationships, simultaneous depiction of vascular and soft-tissue structures, and more display flexibility. However, for many applications, there continues to be controversy in the radiology literature as to the relative efficacy of various rendering techniques. For example, a recent report asserts the practical observation that MIP is superior for small vessels and stenoses, ²³ yet recent experimental data suggests that VR is more accurate for tight stenoses. ²⁴ At our institution, VR is the preferred rendering technique for CTA, with MIP used as an adjunct for smaller vessels within an enhancing organ. ^25,26 CTA rendering techniques should be used in a complementary fashion based on the strengths and limitations of each, though clearly user experience will play a significant role.

Conclusion

Computed tomographic angiography is an established area of diagnostic radiology with rapid improvement and growth driven by technological advances. Conceptual understanding of injection, acquisition, and rendering parameters is critical for the optimization of CTA studies. Much work remains to be done regarding clinically relevant comparison of technical parameter variations for a wide range of specific applications.

Acknowledgments

The author would like to thank Dr. Elliot Fishman for insight and inspiration, and Dr. Leo Lawler for manuscript review.