Note: This copy is for your personal non-commercial use only. To order presentation-ready copies for distribution to your colleagues or clients, contact us at www.rsna.org/rsnarights.

RadioGraphics

CLINICAL APPLICATIONS OF VASCULAR IMAGING

S45

New Techniques in CT Angiography1 CME FEATURE See accompanying test at http:// www.rsna.org /education /rg_cme.html

LEARNING OBJECTIVES FOR TEST 3 After reading this article and taking the test, the reader will be able to: 䡲 Discuss the principles of data acquisition for CT angiography. 䡲 Describe the various postprocessing tools used in neurovascular images. 䡲 List the neurovascular applications of the various image postprocessing tools.

TEACHING POINTS See last page

Michael M. Lell, MD ● Katharina Anders, MD ● Michael Uder, MD Ernst Klotz ● Hendrik Ditt ● Fernando Vega-Higuera, PhD ● Tobias Boskamp, PhD ● Werner A. Bautz, MD ● Bernd F. Tomandl, MD Computed tomographic (CT) angiography has been improved significantly with the introduction of four- to 64-section spiral CT scanners, which offer rapid acquisition of isotropic data sets. A variety of techniques have been proposed for postprocessing of the resulting images. The most widely used techniques are multiplanar reformation (MPR), thin-slab maximum intensity projection, and volume rendering. Sophisticated segmentation algorithms, vessel analysis tools based on a centerline approach, and automatic lumen boundary definition are emerging techniques; bone removal with thresholding or subtraction algorithms has been introduced. These techniques increasingly provide a quality of vessel analysis comparable to that achieved with intraarterial three-dimensional rotational angiography. Neurovascular applications for these various image postprocessing methods include stenoocclusive disease, dural sinus thrombosis, vascular malformations, and cerebral aneurysms. However, one should keep in mind the potential pitfalls of these techniques and always double-check the final results with source or MPR imaging. ©

RSNA, 2006

Abbreviations: ICA ⫽ internal carotid artery, MIP ⫽ maximum intensity projection, MPR ⫽ multiplanar reformation, 3D ⫽ three-dimensional, 2D ⫽ two-dimensional RadioGraphics 2006; 26:S45–S62 ● Published online 10.1148/rg.26si065508 ● Content Codes: 1From

the Department of Radiological Sciences, David Geffen School of Medicine, University of California, Los Angeles (M.M.L.); the Department of Radiology (M.M.L., K.A., M.U., W.A.B.) and Computer Graphics Group (F.V.H.), University of Erlangen-Nuremberg, Maximiliansplatz 1, 91054 Erlangen, Germany; Siemens Medical Solutions, Forchheim, Germany (E.K., H.D.); the MeVis Center for Medical Diagnostic Systems and Visualization, Bremen, Germany (T.B.); and the Department of Neuroradiology, Klinikum-Bremen-Mitte, Bremen, Germany (B.F.T.). Presented as an education exhibit at the 2005 RSNA Annual Meeting. Received February 7, 2006; revision requested April 13 and received May 8; accepted May 17. E.K., H.D., and F.V.H. are employees of Siemens Medical Solutions; T.B. receives research support from Siemens Medical Solutions; all other authors have no financial relationships to disclose. Address correspondence to M.M.L. (e-mail: [email protected]).

©

RSNA, 2006

RadioGraphics

S46

October 2006

Introduction In recent years, rapid advances in computed tomographic (CT) technology and image postprocessing software have been made. CT angiography was improved substantially by increasing scan speed and decreasing section thickness and emerged as a powerful tool in neurovascular imaging. Nowadays, spiral CT systems with acquisition capabilities of up to 64 sections per gantry rotation are introduced in clinical practice. Gantry rotation times decreased to 0.33 second, and section widths of 0.5– 0.6 mm are available. Assessment of vascular studies based on axial images alone is not straightforward; two-dimensional (2D) and three-dimensional (3D) visualization methods are routinely employed to create images comparable to those acquired with catheter angiography. In the emergency situation (stroke or subarachnoid hemorrhage), a robust and fast imaging technique capable of answering all vital clinical questions and allowing clear therapeutic decisions is mandatory. Optimal image quality depends on two factors: CT angiography technique (scan protocol, contrast material injection protocol, image reconstruction methods) and data visualization technique (image postprocessing). The aim of this review is to present optimized data acquisition techniques for multidetector spiral CT and methods of image postprocessing and to discuss their clinical impact in neurovascular imaging. The key question is: Which postprocessing technique is adequate for the clinical question and what are the potential pitfalls?

Technique of Multidetector CT Angiography An essential prerequisite for successful postprocessing is good quality of the acquired imaging data. While the short arteriovenous transit time in neurovascular applications makes short scan times preferable, the small caliber of cervical and intracranial vessels requires the highest spatial resolution in all three dimensions.

Influence of Scan Speed For evaluation of the basal intracranial arteries, a scan range of approximately 100 mm needs to be covered. With four– detector row CT at a collimated section width of 1 mm, a pitch of 1.5, and a gantry rotation time of 0.5 second, this volume can be covered in about 9 seconds. Assuming a cerebral transit time of about 5 seconds, this is not fast enough to avoid venous overlay. With

RG f Volume 26

●

Special Issue

16 – detector row CT at a collimated section width of 0.75 mm, a pitch of 1.5, and a rotation time of 0.5 second, the same range can be covered in 3 seconds, well beyond the arteriovenous transit time. Examination of the whole length of the carotid arteries from the aortic arch to the circle of Willis requires a scan range of approximately 250 mm. With the above-mentioned scan parameters, the scan time would be 21 seconds for four– detector row CT, 7 seconds for 16 – detector row CT, and 4 seconds for 64 – detector row CT (64 ⫻ 0.6 mm, pitch of 1.3, 0.33-second rotation time). The latter protocol allows contrast phase-resolved imaging.

Influence of Spatial Resolution In-plane spatial resolution is predominantly determined by detector geometry and the convolution kernel; it is not substantially improved in scanners with increasing detector row numbers. The major advantage of more detector rows is higher through-plane resolution by reducing the width of a single detector row from 1–1.25 mm (four– detector row CT) to 0.5– 0.6 mm (64 – detector row CT) (1). Typical in-plane resolution with application of a CT angiography protocol (64 ⫻ 0.6-mm detector configuration, 120 kV, 140 mAs [effective], field of view of 120 mm, medium sharp convolution kernel) is 0.6 – 0.7 mm and through-plane resolution is 0.5– 0.7 mm, thus providing isotropic data. Isotropic data allow image reconstruction in arbitrarily chosen planes without loss of spatial resolution and minimization of partial volume effects.

Contrast Material Injection Short scan times require short contrast material injection. The injection protocols need to be simple and standardized to guarantee excellent and reproducible results on a 24-hour basis. To deliver an appropriate amount of iodine, injection rates of 4 –5 mL/sec and highly concentrated contrast medium (iodine, 350 –370 mmol/mL) are preferable. The utility of the contrast material bolus can be increased if a saline bolus is appended. Flushing of the veins reduces streak artifacts due to beam hardening, especially at the thoracic inlet. Individual timing of contrast material injection (bolus tracking or test bolus injection) is mandatory to take advantage of phaseresolved image acquisition. To individualize the timing of contrast material injection, automatic bolus tracking techniques (Smart Prep, CARE Bolus, and Sure Start) can be employed (2). These techniques are fast and easy to use and require only a single contrast material injection. The disadvantage is that a large target vessel for monitoring the contrast material

Teaching Point

●

Special Issue

Lell et al

S47

RadioGraphics

RG f Volume 26

Figure 1. Test bolus method. (a) Axial image shows the right internal carotid artery (ICA) (1), left ICA (2), and left internal jugular vein (3). (b) Diagram shows the enhancement curves for the right ICA (1), left ICA (2), and left internal jugular vein (3) after injection of 10 mL of contrast material and a saline solution bolus. The individual start delay can be set between the arterial peak and the venous upslope. ROI ⫽ region of interest.

arrival is required, and an additional delay for table movement and patient instruction is necessary. Test bolus injection is the alternative to assess the individual circulation time. Its major advantage is that it provides information about the timing of both arterial and venous enhancement in the vessels of interest (Fig 1). The individual start delay can be optimized by placing the scan between the arterial peak and venous contrast material upslope. Table movement and patient instructions can be performed prior to the optimal image acquisition window. The disadvantage is the necessity for an additional injection of about 10 mL of contrast agent (10%–20% increase of total amount).

Image Reconstruction

Teaching Point

To reduce image noise, images may be reconstructed slightly thicker than the detector collimation, for example with a 0.75-mm section thickness from a data set acquired with 64 ⫻ 0.6-mm detector collimation. Overlapping image reconstruction should always be performed to improve 3D postprocessing. The reconstruction increment can be arbitrarily chosen, independent of the detector collimation, but one should keep in mind the amount of resulting data: a reconstruction increment of 50%–75% of the section width may serve as a reasonable rule of thumb. The reconstruction algorithm (convolution kernel) influences the spatial resolution in plane. The ideal kernel would combine low image noise and sharp

edge definition, maintaining good low-contrast resolution. Soft kernels reduce image noise and allow smooth surfaces with rendering techniques, improving the visualization of aneurysms and vascular malformations. Sharper kernels improve edge definition and reduce blooming effects from calcifications, necessary for stenosis measurements, at the expense of higher image noise. The field of view also affects image quality, especially the quality of 3D reformations, which benefit from a small and isotropic voxel size.

Image Postprocessing Techniques Several image processing techniques for CT angiography are currently being used clinically (or at least advertised by the manufacturers). Image processing involves traditional operations such as multiplanar reformation (MPR) and maximum intensity projection (MIP), as well as surface and volume rendering. Because bone and calcifications are seen as a particular problem in CT angiography, a variety of different approaches have been advocated to cope with this problem. Editing, volume cropping, manipulation with transfer functions, and segmentation are common but time-consuming techniques, not convenient in the emergency setting. Visualization with interactive MPR, sliding thin-slab MIP, or standardized volume rendering presets in combination with

RG f Volume 26

October 2006

●

Special Issue

RadioGraphics

S48

Figure 2. Moderate stenosis of the left ICA. (a) Axial source image. (b) Sagittal MPR image. (c) MPR image aligned perpendicular to the vessel optimally depicts the residual lumen (solid arrow) and plaque calcification (dotted arrow).

clip planes is more appropriate. Sophisticated operations like volume rendering with 2D transfer functions or bone subtraction are emerging techniques that enhance the visualization of vascular disease with minimal user interaction.

Multiplanar Reformation

Teaching Point

MPR creates views in arbitrary planes without loss of information. If the CT data meet the requirements of isotropy, spatial resolution is similar to the original source images. Individually adapted image planes exactly perpendicular to the vessel lumen should be created interactively to allow precise quantitative analysis of both lumen and vessel wall. Lumen measurement is dependent on a correct window level setting (3). Both diameter reduction and area reduction can be measured, and no information is suppressed in the final image. No editing is required, but only 2D views can be generated, which may complicate 3D imagination (Fig 2). Therefore, MPR should be applied for precise measurements and be combined with another visualization technique (thin-slab MIP or volume rendering) to display the vessel course and to illustrate where the measurements were performed (4). A variant of MPR is curved planar reformation. Curved planar reformation provides a 2D image that is created by sampling CT volume

data along a predefined curved plane. This technique is employed to display tortuous structures; however, manual definition of curved planes is usually highly error prone and often inappropriate for exact measurements.

Maximum Intensity Projection MIP images are created by displaying only the highest attenuation value from the data encountered by a ray cast through an object to the viewer’s eye (5,6). The depth information along the projection ray is lost; to visualize the spatial relationship of various structures, the volume has to be rotated and viewed from different angles. If bone or calcifications are within the projection ray, these structures are represented on the MIP image instead of the contrast-enhanced vessel because of higher attenuation values. Therefore, bone elimination techniques are essential for pro-

●

Special Issue

RadioGraphics

RG f Volume 26

Lell et al

S49

Surface Rendering

Figure 3. Thin-slab MIP image (slab thickness ⫽ 15 mm) shows the cervical part of the carotid artery. Superimposition of vessels or calcified structures alter lumen visualization (arrow).

Shaded surface display, or surface rendering, is an algorithm that provides a good 3D impression of the surface of an object. In the first step, the surface of an object is separated from other structures. This is usually done by thresholding. Selecting a lower threshold of 200 HU will create a binary data set out of voxels with attenuation values above 200 HU. In the final image, the surfaces of structures meeting this condition will be represented, in CT angiography vessels and bone (Fig 4). As with MIP, bone elimination techniques have to be applied to extract the vascular structures. In the second step, a gray-scale shading procedure is performed to create light intensity in a given 3D scene, simulating surface reflections and shadowing from an artificial light source (8 –10) to enhance depth perception. With binary data, densitometric information gets lost and makes the method prone to undesirable artifacts. Volume rendering has supplanted shaded surface display in virtually all CT angiography indications.

Volume Rendering

Figure 4. Shaded surface display image shows bone and contrast-enhanced vessels as well as calcified plaque. All voxels above the threshold are represented equally. Parts of the jaw were manually removed from the image to exempt the left ICA. The right carotid artery is partly visualized; an occlusion of the right ICA is evident.

cessing vascular MIP images. Superimposition of vessels leads to artificially altered lumen margins, and pathologic conditions may be hidden. To cope with this problem, a modification of MIP called closest vessel projection has been proposed (7). Thin-slab MIP images viewed interactively may be an alternative, as the necessity for bone elimination is limited (Fig 3). MIP is not suitable for the evaluation of stenosis in cases of dense calcification or stents, but thin-slab MIP can provide an excellent “road map” of the vessel course for further evaluation with MPR.

Volume rendering is a visualization technique that creates a 3D impression and provides densitometric information. In volume rendering, all acquired data may be used; therefore, it requires greater processing power than MPR, MIP, or shaded surface display. Visualization of CT angiography data with volume rendering is based on transfer functions that map measured intensities to colors and opacities (11). Opacity values on a spectrum from 0% to 100% (total transparency to total opacity) are assigned along artificial rays that pass through the data (12). Separation of different tissue types (ie, bone, contrast-enhanced vessels, soft tissue) can be performed by applying multiple trapezoids, which can be color encoded. Color can be applied to enhance the discrimination between structures, but color is assigned arbitrarily and does not correlate with the linear progression of gray-scale values on conventional CT images. Decreasing the upslope of the trapezoid is comparable to increasing the window width on a gray-scale image (6). Sliding the trapezoid toward lower Hounsfield unit values on the voxel histogram includes structures with lower attenuation, for example small-caliber vessels, which otherwise

October 2006

RG f Volume 26

●

Special Issue

RadioGraphics

S50

Figure 5. Different transfer function settings alter the representation of the lumen. (a, b) Volume-rendered images created without shading at low opacity (a) and high opacity (b) show accentuated vessel boundaries. (c) On a volume-rendered image created with shading, the 3D impression is improved but edge definition is reduced. (The transfer functions in b and c are identical.) (d) Volume-rendered image created with the transfer function shifted toward higher Hounsfield unit values results in reduced caliber of the visualized vessels.

Teaching Point

might not be classified. The effects of manipulating the trapezoids on the final image can be observed nearly in real time. The definition of the trapezoid strongly affects vascular lumen measurements (Fig 5). Lighting effects enhance the appreciation of spatial relationships between structures. Clip planes are used to remove parts of the volume.

Segmentation Segmentation can be performed manually or (semi)automatically. Segmentation algorithms are often based on the principle of region growing (13). Placing one or more seed points initiates the segmentation of the target structure. From these seed points, more and more neighboring voxels that fulfill predefined criteria are included in the segmentation (14). The technique can be applied in two ways: segmentation of the desired tissue or segmentation of the undesired tissue with subsequent removal from the data. The latter method

●

Special Issue

Lell et al

S51

RadioGraphics

RG f Volume 26

Figure 6. Image created with segmentation shows punched-out defects (arrows) at vessel-bone contact areas. The brachiocephalic vein was removed from the image with additional segmentation; artificial “erosion” of the aortic arch and truncus communis (black patches) resulted from this procedure.

Figure 7. Analysis on the basis of a centerline determined with segmentation and skeletonization. Top left: On a 3D display image, the segmented part of the left carotid artery is colored red. This subvolume was used to create a stretched vessel image (middle) and a crosssectional diagram of vessel diameter (right). On the stretched vessel image, the horizontal structure (arrow) is the external carotid artery; the center of the purple crosshairs is located in the stenosis and indicates the position of the cross-sectional image (bottom left).

removes only interfering tissue (bone or densely enhanced veins) from the CT angiography data and retains soft tissue as well as contrast-enhanced vessels for further evaluation. To refine the boundary of the segmented structures, morphologic dilation operations may be applied. A particular problem in threshold-based segmentation algorithms are areas with close contact of two tissue types with comparable attenuation, such as bone and contrast-enhanced vessels (course of the ICA through the skull base; intraforaminal sections of the vertebral artery) (Fig 6) (15,16). Although the process of segmentation is semiautomatic, user interaction is necessary to set additional seeding points or to intervene in cases of inclusion of neighboring structures due to leakage of the region-growing algorithm. These procedures can be time-consuming and may exceed practical limits in routine clinical work flow.

Vessel Analysis Tools Based on a Centerline Approach Many semiautomatic vessel analysis tools combine the techniques described earlier. The methods for extracting a centerline can be grouped into two categories: (a) One category involves direct computation of an optimal path connecting a given set of points, driven by external factors

(gray value, local contrast) and internal factors (length and curvature) (17,18). (b) In the other method category, the vessel is first segmented with the process of region growing, and the centerline is then determined with a skeletonization process (19 –22). MPR images orthogonal to the vessel path are then computed. The next step is to identify the lumen boundary on these orthogonal cross-sectional images and to perform measurements. To enhance the detection of maximal lumen narrowing and the point of restitution of normal vessel diameter, the vessel can be displayed in a curved planar reformatted image along the centerline (stretched vessel image) and an additional cross-sectional measurement diagram (Fig 7). The cross-sectional measurement diagram represents the diameter values of the selected vessel segment. An exact definition of the centerline is crucial for reliable measurements, because slight deviations from the center can generate artificial stenosis. Interfering factors leading to an inappropriate centerline are calcifications, plaque ulcers, and branching or adjacent vessels (15). A correct

October 2006

RG f Volume 26

●

Special Issue

RadioGraphics

S52

Figure 8. Effect of various threshold values on threshold-based definition of the lumen boundary. Threshold values of 150 HU (a), 200 HU (b), and 250 HU (c) result in calculated stenosis values of 35%, 55%, and 65%, respectively.

lumen boundary definition with exclusion of calcifications is the other prerequisite. Automatic vessel boundary definition depends on explicit or implicit parameter settings; it should be kept in mind that changing these settings may considerably influence the grading of stenosis (Fig 8). Manual adjustment of both the centerline and lumen boundary is prone to individual errors and may be laborious and time-consuming. Because curved planar reformation along a centerline distorts anatomic relationships, the positions of measurements (stenosis and reference site) should be displayed in a 3D scene.

Two-dimensional Transfer Functions As described earlier, standard visualization of CT angiography data with volume rendering is based on transfer functions that map measured intensities to colors and opacities (11). Transfer functions are defined according to the intensities of the structures to be examined. However, if vessels are anatomically not well separated from bone, it is impossible to achieve a clear differentiation. To overcome this limitation, additional features of CT angiography data have to be considered. Tissue interface characteristics in CT data can be described on the basis of Hounsfield unit intensities and their gradient magnitudes (23). A real boundary corresponds to a function that

Figure 9. Profile of the ideal tissue boundary and the corresponding result at CT angiographic reformation. Reformation of CT angiographic data smoothes boundaries to an erf function. The corresponding gradient magnitude reaches its peak at the center of the boundary and decreases at both sides until becoming zero in areas corresponding to uniform tissues.

abruptly jumps between intensity values of neighboring tissues. The limited spatial resolution of CT angiography images does not fully delineate this ideal edge profile; edges are always smoothed to a certain extent. Figure 9 explains the behavior of the gradient magnitude around tissue boundaries for the one-dimensional case. To explore data behavior around 3D boundaries, a 2D histogram that features data intensities and gradient magnitudes is generated. In this structure, parabolic arcs connect the intensity values of adjacent tissues; these parabolic arcs represent the inter-

●

Special Issue

Lell et al

S53

RadioGraphics

RG f Volume 26

Figure 10. (a) Two-dimensional histogram based on intensities (x axis) and gradient magnitudes (y axis) obtained from CT angiographic data. This structure clearly demonstrates tissue boundaries as parabolic arcs. Voxels lying close to tissue boundaries or inside uniform tissue produce histogram “hits” along the upper or lower regions of the parabolas. Open arrow ⫽ air, open arrowhead ⫽ soft tissue, solid arrowhead ⫽ vessels, solid arrow ⫽ osseous tissue. (b) Voxels corresponding to osseous tissue (arrow) and vessels enhanced with contrast medium (arrowhead) are easily identifiable in the 2D transfer function editor. Predefined tissue boundary templates can be interactively placed and adjusted over the corresponding 2D histogram with immediate feedback on volume-rendered images.

complex task of adjusting transfer function thresholds in two dimensions, a predefined transfer function shape that represents the boundary between two tissues was defined (Fig 10b). This parabolic arc object is interactively shifted and sized in order to fit the transfer function to the area representing contrast-enhanced vessels in the histogram with automatic feedback in the volume-rendered images (Fig 11). This technique does not require any kind of preprocessing such as segmentation or filtering and produces highquality results just by adjustment of the applied 2D transfer function.

Bone Subtraction CT Angiography Figure 11. Volume-rendered image obtained after fitting the parabolic arc object to the area representing contrast-enhanced vessels in the 2D histogram. Without further interaction, bone and plaque calcifications are removed from the image.

faces between neighboring anatomic structures (Fig 10a). On the basis of these features, an editor for the creation and adjustment of 2D transfer functions was developed (24). The corresponding 2D histogram is used as background of the working area, so the user gets visual information about voxels belonging to vessels in the CT angiography data. To avoid the

For bone subtraction CT angiography, nonenhanced and contrast-enhanced spiral CT data sets are required. The nonenhanced scan may be a diagnostic scan performed to rule out hemorrhage or ischemia or a low-dose scan performed for subtraction purposes only (16,25). After loading both data sets, processing is performed automatically without any user interaction. The algorithm selectively eliminates bone from the CT angiography data set, retaining both soft tissue and contrast-enhanced vessels. Pixels in the nonenhanced data set with a CT value above

RadioGraphics

S54

October 2006

a certain threshold are defined as bone and used to iteratively register the nonenhanced data set to the CT angiography data set. Registration is rigid (translation and rotation) and based on mutual information (26). After registration, an initial bone mask is defined in the nonenhanced CT volume by means of thresholding. The bone mask is tentatively expanded in three dimensions with morphologic dilation. To reduce the risk of artificial vessel lumen reduction, the optimal contact interface between vessels and bone is determined adaptively. The dilated volume is repeatedly checked for the presence of vessels, and if no vessels are found, the mask is kept expanded and the corresponding voxels are set to a CT value of ⫺1024 HU; otherwise, the corresponding nonenhanced CT voxels are locally subtracted (Fig 12) (16). Bone subtraction CT angiography is a robust method of bone elimination, not requiring user interaction. While patient movement between the two scans can be compensated for in cranial CT angiography, movements can result in incomplete bone or calcification removal in carotid CT angiography, unless additional registration steps or preprocessing is performed. A slab- or sectionbased rigid registration can compensate for movement in the longitudinal axis. A more difficult problem is multidimensional movement of the jaw or the vertebral bones. As rigid registrations cannot capture this complex motion, other techniques were developed. Van Straten et al (27) segmented and registered each bone separately, while Urschler et al (28) automatically separated parts of the volumes that had moved and registered each of these parts separately (Fig 13).

Clinical Applications Carotid Artery Stenosis Atherosclerosis is the most common arterial disease responsible for ischemic stroke. Large trials demonstrated the benefit of carotid endarterectomy in patients with high-grade stenosis (29 – 32). Exact determination of the stenosis is crucial for therapeutic decision making. Atherosclerotic lesions are usually located at the carotid bifurcation, the carotid siphon, and—with smaller hemodynamic relevance—the origin from the aortic arch. Thus, a carotid scan should include the aortic arch as well as the circle of Willis. Pulsation of the aorta and a short arteriovenous circulation time require rapid scanning, favoring 16 – or 64 – detector row CT over four– detector row CT (33–35).

RG f Volume 26

●

Special Issue

Figure 12. MIP image created after bone subtraction CT angiography shows complete elimination of bone; only small calcifications of the hyoid and laryngeal cartilage remain because of swallowing between the nonenhanced and contrast-enhanced acquisitions. The vertebral arteries are clearly demonstrated without artificial lumen reduction at the vertebral foramen. A large lymph node metastasis displaces the left carotid artery; there is mild stenosis of the right ICA.

Atherosclerotic plaques can be grouped into calcified and noncalcified ones. In the presence of calcifications, the residual lumen cannot be assessed with MIP, thin-slab MIP, or surface-rendering techniques without prior bone elimination. Elimination of calcifications with manual image manipulation (editing or segmentation) is both time-consuming and fault prone, therefore unsuitable for daily routine. Semiautomatic segmentation procedures that use threshold-based region-growing techniques can rapidly extract bone or vessels as long as there is a clear separation between both structures (eg, in the cervical part of the carotid arteries). Applying the algorithm on calcified plaque can result in excessive reduction of the residual lumen, which is impossible to check on the final image. Extensive beam-hardening artifacts from metallic dental restorations can interfere with the region-growing algorithm. They may lead to termination of the segmentation process or “algorithmic leakage.” Calcified plaque can be rendered transparent with meticulous parameter setting (Fig 5a). As the parameters of the transfer function significantly affect the displayed lumen diameters, this method is difficult to standardize for accurate and reproducible measurements in different patients and different imaging centers (4). Since algorithms in volume rendering tools are not uniform,

●

Special Issue

Lell et al

S55

RadioGraphics

RG f Volume 26

Figure 13. Volume-rendered (top right) and MIP (bottom right) images show incomplete bone removal due to severe movement between the two acquisitions. Volumerendered (top left) and MIP (bottom left) images created after repetitive registration of subvolumes (28) show optimized bone removal.

Figure 14. CT angiographic images obtained before (a) and after (b) bone subtraction show successful bone removal. However, plaque calcifications (arrow in b) remain in the bone subtraction image because of misregistration due to arterial pulsation.

measurements of the same data may reveal inconsistent results on different workstations. Pulsation of the arteries and soft-tissue shifts at the level of the carotid bifurcation can lead to insufficient removal of calcified plaque when bone mask– based subtraction techniques are used (Fig 14) (16). Attempts are made to correct these

shifts with iterative registration routines and local subtraction (36). If bone mask subtraction is applied to calcified plaque, it is important that only voxels representing bone or calcification are removed from the data without additional mask dilations. Otherwise—as for threshold-based techniques— exaggeration of stenosis may result. As only one scan is employed, movement is irrelevant for 2D transfer function volume rendering. However, luminal representation depends on meticulous parameter setting as well. Manual measurement of the lumen on source or transversely oriented MPR images is easy to standardize and shows excellent interobserver agreement (33,34,37); thus, it should be employed routinely. MPR images of isotropic data provide equivalent spatial resolution as original thin-section (source) images and allow more accurate measurements of vessels not running perpendicular to the scan plane. Commercially available vessel analysis tools implement these procedures. Cross-sectional MPR images perpendicular to the vessel are aligned automatically according to a centerline function. On these cross-sectional images, measurements are performed, and the site of

RadioGraphics

S56

October 2006

RG f Volume 26

●

Special Issue

measurement as well as the results can be superimposed on a volume-rendered image of the corresponding vessel to provide anatomic orientation. Many vessel analysis tools offer automatic lumen measurement procedures. Automated measurements would be desirable in order to obtain reproducible results (Fig 15). Unfortunately, many of the proposed solutions are error prone in cases of branching or nearby passing vessels and may fail in excluding calcifications (15); furthermore, the vessel boundary identification is influenced by either static or adjustable thresholds that have a major impact on stenosis calculation (Fig 8).

Intracranial Stenosis A common location for intracranial atherosclerotic lesions is the carotid siphon. This region is difficult to evaluate with CT angiography, as the vessel is partly embedded in the skull base, with a tortuous course and often circular calcifications of the vessel wall. Orthogonal views of the vessel are required to evaluate the lumen (33), which is most comfortably done by creating a centerline first and reformatting cross-sectional views along this line subsequently with the aid of vessel analysis suites. Adapting the cross-sectional view manually is an alternative, but this may be timeconsuming. Bone subtraction algorithms work well at the skull base because the skull is a rigid compartment and the vessels are sufficiently fixed. These are good conditions for image registration, and complete bone removal as well as removal of calcified plaque can be expected. The exempted vessels can then be visualized with MIP or volume rendering (Fig 16). Segments with a relevant stenosis always need to be reevaluated with adapted cross-sectional images in order to exclude exaggeration of stenosis by local misregistration or an inadequate bone mask. A very convenient solution is to interactively switch between the subtracted and nonsubtracted data sets with identical view settings. The walls of the branches of the intradural cerebral arteries are rarely calcified; therefore, detection of lumen narrowing can be performed with MIP or volume rendering. However, the small size of these vessels makes lumen measurements questionable.

Venous Thrombosis CT venography is a technique employed in the diagnosis of venous thrombosis. The delay be-

Figure 15. High-grade stenosis with circular calcification of the right ICA. Top left: Three-dimensional rendered image highlights the segmented part of the right carotid artery. Cross-sectional MPR image (bottom left) obtained at the current path location, which is indicated by the purple crosshairs on the stretched MPR image (middle), shows the residual lumen surrounded by dense calcification. Right: Cross-sectional diagram shows the results of automatic measurement of area or diameter along the analysis path. The ICA calcification complicates analysis of the residual lumen with automatic and manual procedures.

tween injection of contrast material and data acquisition is targeted to the cerebral veins. In CT venography, scan speed is not a major issue, so image quality does not degrade if four-row scanners are used instead of 16 – 64-section scanners. If bone subtraction CT angiography is applied, a 3D model of the venous cerebral vasculature without interfering bone can be created by using MIP or volume rendering (Fig 17). CT venography has been reported to be accurate in the detection of dural sinus and deep cerebral venous thrombosis (38,39). At CT venography, a thrombosed dural sinus is revealed as an irregular filling defect in the sinus or absence of contrast medium in the sinus (empty delta sign). Before making a diagnosis of cerebral venous thrombosis, anatomic variations of the cerebral veins and dural sinuses should be considered, such as a (unilateral) hypoplastic sinus or sinus fenestration and septa, which may mimic a thrombosed sinus, leading to a false-positive diagnosis (40,41).

●

Special Issue

Lell et al

S57

RadioGraphics

RG f Volume 26

Figure 16. Bilateral stenoses of the distal ICA. (a) MIP image from bone subtraction CT angiography shows the full extents of the stenoses. (b) On a volume-rendered image from CT angiography, parts of the ICAs are hidden. (c) Image from selective catheter angiography shows the same findings as CT angiography. (The image was created from two digital subtraction angiographic series.)

Vascular Malformations

Figure 17. Tumor invasion of the right transverse sinus. MIP image from bone subtraction CT venography shows the large cerebral veins and sinuses. Because only the voxels representing bone are removed from the image, the soft tissues remain for further evaluation.

In cases of sinus thrombosis, small thrombi may be undetected with surface rendering and MIP due to collateral contrast material flow or dural enhancement. Once again, interactive thinslab MIP or MPR is preferred.

Three-dimensional rendering of vascular malformations in the head and neck provides an excellent anatomic overview of the lesion and may enhance spatial perception for the surgeon or interventional radiologist. CT angiography is not suited to exclusion of small intracranial arteriovenous malformations; delineation of the nidus is another domain of digital subtraction angiography, which provides the highest spatial and temporal resolution as well as dynamic information (42). In extracranial malformations and hemangiomas, CT can demonstrate both the lesion and the surrounding tissue, information that is critical for therapy planning. Interactive MPR and thinslab MIP are suited to analysis of feeding and draining vessels, but the 3D presentation is limited. Bone suppression (segmentation, 2D transfer function, bone subtraction CT angiography) may be advantageous if the lesion is partly embedded in bony structures (Fig 18). Rapid scanning is needed to differentiate the arterial and venous sides, depending on the size of the lesion, although in high-flow arteriovenous malformations this may be impossible even with the latestgeneration scanners.

RadioGraphics

S58

October 2006

Figure 18. Large vascular malformation with significant arteriovenous shunting. (a, b) Coronal MPR (a) and thinslab MIP (b) images show the internal structure of the lesion and thinning of the skull in detail. (c, d) Volumerendered images created with the one-dimensional transfer function technique (c) and from segmented data with a high-opacity setting (d) provide the best 3D representation but do not show the thrombosed parts of the lesion. (e, f) Volume-rendered image from bone subtraction CT angiography (e) and image from digital subtraction angiography (f) show that the lesion has no feeding vessels from the ICA (inset).

RG f Volume 26

●

Special Issue

●

Special Issue

Lell et al

S59

RadioGraphics

RG f Volume 26

Figure 19. Aneurysms of the right ICA and left posterior cerebral artery. (a) On an image created with one-dimensional transfer functions, vessels and bone cannot be well differentiated because of an overlap in the attenuations of these structures. (b) Volume-rendered image from bone subtraction CT angiography shows the vessels clearly. Use of a high-opacity setting improves the 3D representation; however, the enhanced cavernous sinus hides small portions of the ICA. (c) On an image created with a low-opacity setting, the sinus is transparent, thus allowing visualization of the vessel boundary. (d) Volume-rendered image created with 2D transfer functions shows similar results.

Aneurysms Mortality is high among patients with aneurysm rupture, and prompt localization of the aneurysm is critical to determine the appropriate neurosurgical or endovascular intervention. Some authors advocate multidetector CT angiography as the primary method to evaluate cerebral aneurysms (43,44). The short scan times of 64 – detector row scanners allow arterial phase imaging with clearly different attenuation values of arteries and veins, while marked venous enhancement is likely to occur with four– to 16 – detector row CT. Venous enhancement is not a crucial factor in the detection of cerebral aneurysms, except for extensive enhancement of the cavernous sinus. Venous structures may be removed with segmentation tools. Supraclinoid aneurysms can be well visualized by applying MIP or volume rendering (16,44 – 46); volume rendering provides the best 3D impression. Interactive manipulation of the volume rendering presets is necessary to differentiate be-

tween infundibular dilatation of the vessel origin and true aneurysm (46). Detection of aneurysms located beyond the clinoid process is more difficult (47) because bony structures partly obscure the vessels. Bone suppression (2D transfer function volume rendering) or bone subtraction techniques improve the delineation of infraclinoid aneurysms, providing free access to the vessel in question (Fig 19) (16,45,48). Follow-up with CT angiography after interventional or surgical treatment faces considerable challenges: Clips or coils usually cause beamhardening artifacts, altering the Hounsfield unit values in surrounding soft tissue and vessels. If volume rendering techniques are applied, the affected vessel segment may not be represented by the transfer function, generating the impression of vascular stenosis or occlusion. Bone subtraction CT angiography may completely remove coils or

Teaching Point

RG f Volume 26

October 2006

●

Special Issue

RadioGraphics

S60

Figure 20. Follow-up after clipping of an aneurysm. (a) Volume-rendered image from bone subtraction CT angiography (view from above) shows a simulated occlusion of the right distal ICA (C7) and proximal anterior (A1) and medial (M1) cerebral arteries. (b) Image created from the original CT angiographic data shows the location of the aneurysm clip, which was completely removed from the image. (c) Thin-slab MIP image shows the clip and beam-hardening artifacts.

Postprocessing Strategies for Various Neurovascular Indications Neurovascular Indications Stenosis Malformation or aneurysm Infraclinoid or close bone contact Supraclinoid Thrombosis

Postprocessing Strategies* MPR plus centerline analysis or MPR plus thin-slab MIP and VR (scout) Bone subtraction CT angiography plus VR or 2D TF VR VR MPR or thin-slab MIP

Note.—The results should always be double-checked with interactive MPR imaging. *TF ⫽ transfer function, VR ⫽ volume rendering.

clips in the final data set, hampering the identification of this artifact (Fig 20). A summary of the postprocessing strategies for the clinical indications is given in the Table.

Conclusions In neurovascular imaging, selective catheter angiography is still considered the standard of refer-

ence, but multidetector CT is increasingly used as a noninvasive alternative. Contrast phase-resolved CT of vascular lesions in the head and neck can be performed in a couple of seconds, providing angiographic information as well as information on the surrounding soft tissue. Generating “boneless” 3D images became possible with modern postprocessing techniques, but one should keep in mind the potential pitfalls of these techniques and always double-check the final results with source or MPR images.

RadioGraphics

RG f Volume 26

●

Special Issue

References 1. Flohr TG, Stierstorfer K, Ulzheimer S, Bruder H, Primak AN, McCollough CH. Image reconstruction and image quality evaluation for a 64-slice CT scanner with z-flying focal spot. Med Phys 2005; 32:2536 –2547. 2. Napoli A, Fleischmann D, Chan FP, et al. Computed tomography angiography: state-of-the-art imaging using multidetector-row technology. J Comput Assist Tomogr 2004;28:S32–S45. 3. Liu Y, Hopper KD, Mauger DT, Addis KA. CT angiographic measurement of the carotid artery: optimizing visualization by manipulating window and level settings and contrast material attenuation. Radiology 2000;217:494 –500. 4. Luccichenti G, Cademartiri F, Pezzella FR, et al. 3D reconstruction techniques made easy: knowhow and pictures. Eur Radiol 2005;15:2146 – 2156. 5. Napel S, Marks MP, Rubin GD, et al. CT angiography with spiral CT and maximum intensity projection. Radiology 1992;185:607– 610. 6. Calhoun PS, Kuszyk BS, Heath DG, Carley JC, Fishman EK. Three-dimensional volume rendering of spiral CT data: theory and method. RadioGraphics 1999;19:745–764. 7. Zuiderveld K. Visualisation of multimodality medical volume data using objectoriented methods [doctoral dissertation]. Utrecht, the Netherlands: Utrecht University, 1995. 8. Magnusson M, Lenz R, Danielsson PE. Evaluation of methods for shaded surface display of CT volumes. Comput Med Imaging Graph 1991;15: 247–256. 9. Rubin GD, Dake MD, Napel SA, McDonnell CH, Jeffrey RB Jr. Three-dimensional spiral CT angiography of the abdomen: initial clinical experience. Radiology 1993;186:147–152. 10. Rubin GD, Dake MD, Napel S, et al. Spiral CT of renal artery stenosis: comparison of three-dimensional rendering techniques. Radiology 1994;190: 181–189. 11. Levoy M. Display of surfaces from volume data. IEEE Comput Graph Appl 1988;8:29 –37. 12. Dalrymple NC, Prasad SR, Freckleton MW, Chintapalli KN. Introduction to the language of three-dimensional imaging with multidetector CT. RadioGraphics 2005;25:1409 –1428. 13. Blum H. Biological shape and visual science. I. J Theor Biol 1973;38:205–287. 14. Udupa JK. Three-dimensional visualization and analysis methodologies: a current perspective. RadioGraphics 1999;19:783– 806. 15. Zhang Z, Berg MH, Ikonen AE, Vanninen RL, Manninen HI. Carotid artery stenosis: reproducibility of automated 3D CT angiography analysis method. Eur Radiol 2004;14:665– 672. 16. Lell M, Anders K, Klotz E, Ditt H, Bautz W, Tomandl BF. Clinical evaluation of bone-subtraction CT angiography (BSCTA) in head and neck imaging. Eur Radiol 2006;16:889 – 897. 17. Hernandez-Hoyos M, Orkisz M, Puech P, Mansard-Desbleds C, Douek P, Magnin IE. Computer-assisted analysis of three-dimensional MR angiograms. RadioGraphics 2002;22:421– 436.

Lell et al

S61

18. Frangi AF, Niessen WJ, Hoogeveen RM, van Walsum T, Viergever MA. Model-based quantitation of 3-D magnetic resonance angiographic images. IEEE Trans Med Imaging 1999;18:946 –956. 19. Boskamp T, Rinck D, Link F, Kummerlen B, Stamm G, Mildenberger P. New vessel analysis tool for morphometric quantification and visualization of vessels in CT and MR imaging data sets. RadioGraphics 2004;24:287–297. 20. Borgefors G, Ramella G, Sanniti di Baja G. Hierarchical decomposition of multiscale skeletons. IEEE Trans Pattern Anal Mach Intell 2001;23: 1296 –1312. 21. Boskamp T, Hahn HK, Hindennach M, et al. Geometrical and structural analysis of vessel systems in 3D medical image datasets. In: Leondes CT, ed. Medical imaging systems technology. Singapore: World Scientific, 2005. 22. Lam L, Lee SW, Suen CY. Thinning methodologies: a comprehensive survey. IEEE Trans Pattern Anal Mach Intell 1992;14:869 – 885. 23. Kindlmann G, Durkin JW. Semi-automatic generation of transfer functions for direct volume rendering. In: Proceedings of the 1998 IEEE symposium on volume visualization. New York, NY: ACM Press, 1998; 79 – 86. 24. Vega Higuera F, Sauber N, Tomandl BF, Nimsky C, Greiner G, Hastreiter P. Enhanced 3D-visualization of intracranial aneurysms involving the skull base. Lect Notes Comput Sci 2003;2879: 256 –263. 25. Venema HW, Hulsmans FJH, den Heeten GJ. CT angiography of the circle of Willis and intracranial internal carotid arteries: maximum intensity projection with matched mask bone elimination— feasibility study. Radiology 2001;218:893– 898. 26. Pluim JP, Maintz JB, Viergever MA. Mutual-information-based registration of medical images: a survey. IEEE Trans Med Imaging 2003;22:986 – 1004. 27. van Straten M, Venema HW, Streekstra GJ, Majoie CB, den Heeten GJ, Grimbergen CA. Removal of bone in CT angiography of the cervical arteries by piecewise matched mask bone elimination. Med Phys 2004;31:2924 –2933. 28. Urschler M, Ditt H, Bischof H. Partially rigid bone registration in CT angiography. Presented at the 11th Computer Vision Winter Workshop, Telc, Czech Republic, February 6 – 8, 2006. 29. Executive Committee for the Asymptomatic Carotid Atherosclerosis Study. Endarterectomy for asymptomatic carotid artery stenosis. JAMA 1995; 273:1421–1428. 30. Randomised trial of endarterectomy for recently symptomatic carotid stenosis: final results of the MRC European Carotid Surgery Trial (ECST). Lancet 1998;351:1379 –1387. 31. Barnett HJ, Taylor DW, Eliasziw M, et al. Benefit of carotid endarterectomy in patients with symptomatic moderate or severe stenosis. North American Symptomatic Carotid Endarterectomy Trial Collaborators. N Engl J Med 1998;339:1415– 1425.

RadioGraphics

S62

October 2006

32. Rothwell PM, Eliasziw M, Gutnikov SA, et al. Analysis of pooled data from the randomised controlled trials of endarterectomy for symptomatic carotid stenosis. Lancet 2003;361:107–116. 33. Bash S, Villablanca JP, Jahan R, et al. Intracranial vascular stenosis and occlusive disease: evaluation with CT angiography, MR angiography, and digital subtraction angiography. AJNR Am J Neuroradiol 2005;26:1012–1021. 34. Bartlett ES, Walters TD, Symons SP, Fox AJ. Quantification of carotid stenosis on CT angiography. AJNR Am J Neuroradiol 2006;27:13–19. 35. Lell M, Wildberger JE, Heuschmid M, et al. CTangiography of the carotid artery: first results with a novel 16-slice-spiral-CT scanner [in German]. Rofo 2002;174:1165–1169. 36. van Straten M, Venema HW, Streekstra GJ, Reekers JA, den Heeten GJ, Grimbergen CA. Removal of arterial wall calcifications in CT angiography by local subtraction. Med Phys 2003;30:761–770. 37. Hirai T, Korogi Y, Ono K, et al. Maximum stenosis of extracranial internal carotid artery: effect of luminal morphology on stenosis measurement by using CT angiography and conventional DSA. Radiology 2001;221:802– 809. 38. Ozsvath RR, Casey SO, Lustrin ES, Alberico RA, Hassankhani A, Patel M. Cerebral venography: comparison of CT and MR projection venography. AJR Am J Roentgenol 1997;169:1699 –1707. 39. Casey SO, Alberico RA, Patel M, et al. Cerebral CT venography. Radiology 1996;198:163–170. 40. Provenzale JM, Joseph GJ, Barboriak DP. Dural sinus thrombosis: findings on CT and MR imaging and diagnostic pitfalls. AJR Am J Roentgenol 1998;170:777–783.

RG f Volume 26

●

Special Issue

41. Ayanzen RH, Bird CR, Keller PJ, McCully FJ, Theobald MR, Heiserman JE. Cerebral MR venography: normal anatomy and potential diagnostic pitfalls. AJNR Am J Neuroradiol 2000;21: 74 –78. 42. Zhang XQ, Shirato H, Aoyama H, et al. Clinical significance of 3D reconstruction of arteriovenous malformation using digital subtraction angiography and its modification with CT information in stereotactic radiosurgery. Int J Radiat Oncol Biol Phys 2003;57:1392–1399. 43. Tipper G, U-King-Im JM, Price SJ, et al. Detection and evaluation of intracranial aneurysms with 16-row multislice CT angiography. Clin Radiol 2005;60:565–572. 44. Villablanca JP, Jahan R, Hooshi P, et al. Detection and characterization of very small cerebral aneurysms by using 2D and 3D helical CT angiography. AJNR Am J Neuroradiol 2002;23:1187– 1198. 45. Tomandl BF, Hammen T, Klotz E, Ditt H, Stemper B, Lell M. Bone-subtraction CT angiography for the evaluation of intracranial aneurysms. AJNR Am J Neuroradiol 2006;27:55–59. 46. Tomandl BF, Kostner NC, Schempershofe M, et al. CT angiography of intracranial aneurysms: a focus on postprocessing. RadioGraphics 2004;24: 637– 655. 47. Tomandl BF, Hastreiter P, Iserhardt-Bauer S, et al. Standardized evaluation of CT angiography with remote generation of 3D video sequences for the detection of intracranial aneurysms. RadioGraphics 2003;23:e12. 48. Matsumoto M, Kodama N, Sakuma J, et al. 3D-CT arteriography and 3D-CT venography: the separate demonstration of arterial-phase and venous-phase on 3D-CT angiography in a single procedure. AJNR Am J Neuroradiol 2005;26:635– 641.

This article meets the criteria for 1.0 credit hour in category 1 of the AMA Physician’s Recognition Award. To obtain credit, see accompanying test at http://www.rsna.org/education/rg_cme.html.

RadioGraphics

RG

Volume 26 • Special Issue • October 2006

Lell et al

New Techniques in CT Angiography Michael M. Lell, MD, et al RadioGraphics 2006; 26:S45–S62 ● Published online 10.1148/rg.26si065508 ● Content Codes:

Page S46 Individual timing of contrast material injection (bolus tracking or test bolus injection) is mandatory to take advantage of phase-resolved image acquisition. Page S47 The reconstruction increment can be arbitrarily chosen, independent of the detector collimation, but one should keep in mind the amount of resulting data: a reconstruction increment of 50%–75% of the section width may serve as a reasonable rule of thumb. Page S48 Therefore, MPR should be applied for precise measurements and be combined with another visualization technique (thin-slab MIP or volume rendering) to display the vessel course and to illustrate where the measurements were performed (4). Page S50 The definition of the trapezoid strongly affects vascular lumen measurements (Fig 5). Pages S59 Bone suppression (2D transfer function volume rendering) or bone subtraction techniques improve the delineation of infraclinoid aneurysms, providing free access to the vessel in question (Fig 19) (16,45,48).