Dr. Ravenel is an Assistant Professor of Thoracic Imaging in the Department of Radiology at the Medical University of South Carolina, Charleston, SC. Dr. McAdams is an Associate Professor of Radiology and Dr. Goodman is a Professor of Radiology and Section Chief of Thoracic Imaging at Duke University Medical Center, Durham, NC

Pulmonary embolism (PE) is a frequently suspected disease that has a confusing clinical picture and an uncertain outcome. It has been stated that PE is a major contributor to death in large numbers of patients. ¹ PE is said to be the sole cause or major contributor to death in up to 37% of patients in autopsy studies where PE was present. ² Furthermore, it has been stated that the untreated mortality rate from PE may be as high as 30% ³ and that anticoagulation therapy may reduce this mortality rate to <3%. ⁴ These data are, however, controversial. Some investigators believe that PE is overdiagnosed and that its significance is overstated. ^5-8

Previously, ventilation-perfusion scintigraphy and pulmonary angiography were the imaging tests of choice in cases of suspected PE. Due to limitations in the accuracy of ventilation-perfusion (V/Q) scanning and the invasive nature and perceived complications of angiography, coupled with improvements in spiral computed tomography (CT), a great deal of interest in using spiral CT for screening and diagnosis of PE has been generated. ⁹ Since 1996, a number of studies evaluating spiral CT for the diagnosis of PE have been performed, with a wide range of sensitivities (53% to 92%) and specificities (78% to 100%) reported. ^10-13 However, in outcome-based studies, the negative predictive value for CT approaches 100%. That is, a normal CT correlates with extremely low rates of morbidity and mortality due to subsequent PE. ^14,15 While pulmonary angiography is still considered by many to be the gold standard for diagnosis of PE, CT ultimately may achieve this status. ^16,17

The implementation of CT into the diagnostic algorithm for PE at hospitals has been variable and is related to the individual clinician's acceptance of the technique as well as the radiologist's skill in interpretation. The implication of false-negative interpretations would be great if this occurred with significant frequency. However, as mentioned previously, the negative predictive value of a normal CT is virtually as high as a normal V/Q scan and normal pulmonary angiogram. Perhaps more important are false-positive diagnoses, which can also have significant consequences. Oral anti-coagulation may result in fatal and major bleeding in 0.25% to 1% and 1% to 5% of patients respectively, ¹⁸ and thrombolysis has a major bleeding incidence of up to 20%. ¹⁹ Fortunately, CT scanning is more accurate than V/Q scanning ^11,20 and should alleviate this problem in many patients.

Nevertheless, in order to gain acceptance, both positive and negative diagnoses must be made confidently. Our goal is to review and demonstrate the various diagnostic pitfalls and artifacts that can complicate the interpretation of CT pulmonary angiography.

Imaging techniques

Techniques for spiral CT pulmonary angiography have been described previously ^21-23 and vary from institution to institution. The most important parameters are: 1) rapid bolus infusion of intravenous (IV) contrast material through a large IV catheter; 2) appropriate scan delay to optimize opacification of the pulmonary arteries; and 3) thin collimation (1 to 3 mm). Representative protocols are listed in Table 1.

In our practice, we typically infuse 150 cc of nonionic contrast at a rate of 4 cc/second via power injector. This rapid infusion requires placement of a large bore IV catheter, preferably 18-gauge, in a large vein. We usually require placement in an antecubital vein. We will not power inject through a small-gauge peripheral catheter (>20-gauge), a central venous catheter, a peripherally inserted central catheter (PICC), or a catheter placed in a more superficial (hand) vein.

In the majority of patients, peak enhancement in the pulmonary arteries occurs between 15 and 25 seconds after beginning the injection. A number of medical conditions, particularly congestive heart failure, can delay pulmonary artery opacification. There are three methods for optimizing pulmonary arterial opacification. Some institutions, including ours, use a standard timing delay . We use a 25-second delay in most patients; if the patient has known cardiac disease, we increase the delay by 5 seconds. We have found that this approach works very well in our institution. Bolus tracking is a second method for optimizing pulmonary arterial opacification. This technique is available on most CT scanners and is known by a variety of proprietary terms (for example, "Smart Prep" on General Electric scanners [GE Medical Systems, Milwaukee, WI]). Using this technique, the patient is repeatedly scanned at low
milliampere (mA) at a single preselected level, usually the main pulmonary artery. A region of interest is placed on the main pulmonary artery and the scan is initiated at either a preset level of opacification or at peak contrast opacification. This technique is particularly useful when injecting from the lower extremities. Bolus timing is a third method for optimizing pulmonary arterial opacification. Using this method, a small bolus (20 to 30 cc) of IV contrast is injected, serial images are obtained over the main pulmonary artery, and the time to peak opacification is determined. The scan delay is then set at this time plus a few seconds (typically 3 seconds). This technique works quite well, but it is more time-consuming to setup than the standard delay technique. The choice of technique used in an individual practice is dictated by a variety of concerns, including type of CT scanner, experience of CT technologists, and patient population.

Choice of slice collimation and pitch depends on a number of factors, including type of machine (manufacturer, single versus multidetector, rotation speed), area to be covered and length of breath-hold. Thin-collimation is absolutely essential for confident exclusion of PE. On a single detector scanner, collimation should be no greater than 3 mm; on a multidetector scanner, thinner collimation is both possible and desirable. We perform all of our PE studies on a multidetector scanner and typically use 2.5-mm collimation with 15-mm/rotation table speed, 0.8-second rotation. Some have found that 1.25-mm collimation with 7.5-mm/rotation table speed improves depiction of subsegmental vessels. Whether or not this improves detection of small emboli is not known. We currently use 2.5-mm collimation because it allows us to complete the scan in <10 seconds--a very reasonable breath-hold for all but the most acutely ill patients. As CT scanners are upgraded to faster rotation speeds (from 0.8 seconds to 0.5 seconds or less), it may be feasible to acquire the studies in <10 seconds using 1.25-mm collimation.

Axial images should be reconstructed at intervals no greater than, and preferably less than, slice collimation. Overlapping reconstructions helps eliminate stair-step artifacts (see below) and facilitates multiplanar reconstructions. We typically reconstruct the axial images at 1-mm intervals. This results in a large number of images per case (150 to 250), necessitating review of the study on a workstation. We find it difficult and more time-consuming to read the studies in hard-copy format. Scrolling through individual images in stepwise fashion on a workstation makes it easier to follow vessels travelling perpendicular to the axial plane. Furthermore, two-dimensional (2D) images reconstructed in coronal, sagittal, or oblique projections can be quite useful for assessing vessels that run either parallel or slightly oblique to the axial plane. ²⁴

Diagnostic pitfalls

A variety of normal structures and related conditions may complicate the interpretation of CT scans performed to evaluate for PE. The usual appearance of PE on CT studies is a central filing defect in a pulmonary artery surrounded by a rim of contrast, complete cut-off of a vessel (figure 1), a serpiginous low attenuation clot in larger vessels (figure 2), or, exceptionally, as high attenuation thrombus on a non-contrast CT (figure 3). Venous structures, lymphoid and perivascular tissue, vessel orientation, and mucus or fluid in adjacent bronchi all can lead to false-positive interpretations. Occasionally, chronic pulmonary embolism may be confused with acute disease and rarely nonthrombotic emboli may be seen macroscopically. In this section, we will discuss the various normal and pathologic conditions that can mimic PE on spiral CT scan.

Venous structures

Since scanning parameters are designed for optimal arterial enhancement, enhancement of the pulmonary veins is quite variable. Flow artifacts in poorly filled veins may simulate PE (figure 4A). It is, therefore, important not to mistake a poorly filled vein for an artery. Owing to their proximity to the central hilar vessels, the superior pulmonary veins can be mistaken for arteries easily. The right and left superior pulmonary veins descend anterior to the left and right upper lobe pulmonary arteries at the bifurcation of the anterior segmental artery (A2 segment). With careful viewing, the veins can be traced into the left atrium. As a general rule, arteries and bronchi run in parallel, while the veins return to the left atrium individually. In cases where there is confusion, 2D coronal and sagittal reformatted images (figure 4B) and viewing images at lung window settings can be helpful (figure 5).

Lymphoid tissue

Normal lymphoid tissue can present difficulty in interpretation, particularly at arterial branch points. Normal lymph nodes are frequently present lateral to the A2 segment of the upper lobes (figure 6), medial to the interlobar and lower lobe arteries, and at take-off of the lower lobe segmental arteries. ²⁵ The distinction can be more difficult in patients with neoplasm where either tumor or lymph node metastases may distort the normal arterial anatomy. Acute emboli often are surrounded by a rim of contrast that can distinguish them from lymphoid tissue. At bifurcation points, emboli may impact and not be completely surrounded by contrast. In these cases, attention to vessel caliber and a careful search for distal emboli often lead to the correct interpretation.

Vessel orientation

The majority of upper and lower lobe vessels run in a superior/inferior plane. Difficulty in tracing middle lobe and lingular arteries exists as they generally run oblique to the axial scan plane. Partial volume averaging can lead to the appearance of low attenuation filling defect within these vessels (figure 7), and as every hard-copy image may not be printed, attention to these regions on the monitor is essential. Obliquely oriented vessels can often be better evaluated with coronal and sagittal reformatted images.

Mucus or fluid in bronchi

Patients with chronic obstructive pulmonary disease, those that are ventilator-dependent, and postoperative patients all may have difficulty clearing secretions. This may lead to the accumulation of mucus or fluid in segmental bronchi. On axial images, fluid-filled bronchi have a low attenuation center with a rim of higher attenuation (bronchial wall), mimicking the appearance of PE (figure 8). Because the arteries and bronchi run parallel to each other, a contrast-filled vessel should be apparent immediately adjacent to the fluid-filled bronchus. Reformatted images can also better depict the anatomy and differentiate fluid-filled airways from PE.

Nonthrombotic emboli

It is rare for nonthrombotic emboli to present with macroscopic filling defects. Neoplasms that grow into the inferior vena cava (renal, adrenal, and hepatic) and atrial myxomas can fragment and present as visible filling defects. Because of similar Hounsfield unit (HU) measurements, these are indistinguishable from thrombotic emboli. The possibility of nonthrombotic emboli should be considered in the appropriate clinical setting. ²⁶ Exceptionally, macroscopic fat emboli can be encountered and diagnosed by HUs consistent with fat.

Chronic versus acute embolus

Chronic PE may be present particularly in those with long-standing deep venous thrombosis. The distinction is difficult in the segmental and subsegmental arteries. In the main and lobar arteries, chronic PE is applied closely to the vessel wall and has obtuse angles at its periphery (figure 9). Abrupt vessel cut-off, intravascular webs, intimal irregularity, and pouch-like defects may also be seen. ²⁷

Imaging artifacts

Artifacts in CT imaging come from inherent problems with the imaging system (beam hardening from high-density contrast), image reconstruction (stair-step artifact) and patient cooperation (breathing and motion artifact). All of these artifacts can make studies difficult to interpret.

Artifacts of enhancement

Despite rapid infusion of contrast and a 20 to 30-second scan delay, there is often marked enhancement of the superior vena cava during the scan. Dense opacification of the superior vena cava can cause beam hardening and streak artifact obscuring adjacent vessels (figure 10). In order to minimize the artifact, some authors suggest diluting the contrast. In one report, a 1:1 dilution of
300 mg/mL iodine results in less artifact compared to undiluted contrast; however, further dilution results in unacceptable arterial enhancement. ²⁸ In our experience, this artifact does not significantly impact the quality of the images and we do not typically dilute the contrast.

A poor contrast bolus can similarly make interpretation of CT scans for PE difficult (figure 10). The detection of PE relies on HU differences between PE and the contrast bolus. Poor opacification of the pulmonary artery decreases the attenuation differences between the contrast bolus and a PE. At wider window settings, the embolism may not be distinguished from contrast within the vessel. Viewing at narrower window settings can help accentuate subtle differences in film contrast, but in cases where a firm diagnosis must be established, a repeat contrast-enhanced CT scan or other imaging test (V/Q scan or conventional pulmonary angiography) should be performed.

Stair-step artifact

Stair-step artifacts result from data reconstruction and misregistration. With loss of data along the z-axis, apparent areas of both decreased and increased attenuation occur. This apparent decrease in attenuation of the vessel lumen on multiple images can mimic the appearance of PE. On axial images, the vessel lumen is generally affected on alternating images with normal intervening vessel enhancement. On reconstructed images, the "stair-step" appearance can be appreciated (figure 11). Decreasing the slice thickness and overlapping images will reduce the amount of artifact along the z-axis; however, when slice thickness is too thin, quantum mottle, scan time, and radiation dose to the patient can provide significant problems.

Breathing artifact

Patients with suspected PE are usually short of breath. With a 24-second scan time on single-slice scanners, the patient often cannot perform a full breath-hold. This is less of a problem with multislice scanning due to the faster scan time. Breathing artifacts cause blurring of the image (figure 12), making assessment of the vessel lumen impossible. While the ultimate solution is decreasing scan time, scanning from bottom to top allows images to be obtained at the lung bases first, where motion and breathing artifact will be most pronounced. More superiorly, breathing has less of an effect and blurring of vessels is generally not a problem.

Conclusion

The success of CT in the imaging of PE not only depends on the interpreter, but also the quality of the technique. A knowledge of the various pitfalls and artifacts allows for a more confident interpretation of positive and negative studies as well as fewer false-positive and false-negative examinations. AR