Common Artifacts in PET Myocardial Perfusion Images Due to Attenuation–Emission Misregistration: Clinical Significance, Causes, and Solutions

Study Patients

Between May 2000 and July 2003, a total of 1,177 consecutive diagnostic myocardial perfusion rest-dipyridamole PET studies were performed at the University of Texas Weatherhead PET Center for Preventing and Reversing Atherosclerosis. An additional 45 patients underwent cardiac PET imaging with 3 sequential resting attenuation scans to assess diaphragmatic changes on assuming the supine position. All subjects signed informed consent approved by the Committee for the Protection of Human Subjects of the University of Texas Health Science Center, Houston.

PET Imaging

Patients were instructed to fast for 4 h and abstain from caffeine, theophylline, and cigarettes for 24 h before study. As previously described (4–6,8), PET imaging was performed using the University of Texas–designed, Positron Posicam Auricle, bismuth germanate, 2-dimensional (2D) multislice tomograph with a reconstructed resolution of 10-mm full width at half maximum (FWHM). Based on a 5-min positioning transmission scan, patients were precisely positioned in the PET scanner with the first image plane 1 cm below the radiolucent bronchial carina on the positioning transmission scan. Laser guides aligned to external body markers were used to check correct position at least 3 times throughout data acquisition with adjustments before each image acquisition to maintain precise alignment according to the external body markers.

Using a rotating rod source containing 148–185 MBq (4–5 mCi) of ⁶⁸Ge, transmission images to correct for photon attenuation contained 100–150 million counts; the transmission scan image is also called the attenuation image in this article. Emission images obtained after intravenous injection of 925 MBq to 2.22 GBq (25–60 mCi) of generator-produced ⁸²Rb contained 40–60 million counts or after 666 MBq (18 mCi) of cyclotron-produced ¹³N contained 20–40 million counts.

Immediately after completing resting ⁸²Rb or 30 min after administration of the first dose of ammonia, dipyridamole (0.142 mg/kg/min) was infused for 4 min. At 4 min after completion of the dipyridamole infusion, the same dose of the same radionuclide was given intravenously. For ⁸²Rb, imaging data acquisition was started at 60 s (normally) to 90 s (for patients with heart failure or cardiac enlargement) after the beginning of ⁸²Rb infusion. For ¹³N-ammonia, imaging data acquisition was started at 4 min after injection to allow blood-pool clearing needed for lung washout of ¹³N-ammonia. PET imaging was repeated by the same protocol as for the resting study. For angina, aminophylline (125 mg) was given intravenously.

Automated Quantitative Analysis of PET Images

Images were reconstructed using filtered backprojection with a Butterworth filter having a cutoff of 0.4 and roll off of 10. Completely automated analysis of severity and size of PET abnormalities was performed by previously described software (1–8). A 3D restructuring algorithm generates true short- and long-axis views from PET transaxial cardiac images, perpendicular to and parallel to the long axis of the left ventricle. From the tomographic data acquired in 2D mode to minimize scatter, circumferential profiles are used to reconstruct 3D topographic views of the left ventricle showing relative regional activity distribution divided into lateral, inferior, septal, and anterior quadrant views of the 3D topographic display corresponding to the coronary arteries as previously described (3–8) and illustrated in Figure 1A.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1.

Severe rest perfusion defect. (A) Topographic display. Top panel shows topographic display of unshifted images with artifactual defect due to misregistration of emission and attenuation data. Middle panel shows topographic display of same patient after shifting emission images for optimal coregistration with attenuation image. Bottom panel shows our myocardial perfusion map relating 3D quadrant views of PET images to arterographic anatomy from (4). RI = ramus intermedius; OM₁ = first obtuse marginal branch; OM₂ = second obtuse marginal branch; LCx = left circumflex artery; D₁ = first diagonal branch; D₂ = second diagonal branch; RCA = right coronary artery; LAD = left anterior descending (artery). (B) Quantitative analysis of misregistration. Top panel shows rotated transmission images in scanner horizontal longitudinal section before shifting (left) and after shifting emission data (right) with superimposed outline of emission image in same tomographic slice using same rotation angles. Middle panel shows rotated transmission images in scanner vertical longitudinal section before shifting (left) and after shifting emission data (right) with superimposed outline of emission image in same tomographic slice using same rotation angles. Bottom panel shows transmission images (red color) in in-plane acquisition tomographic view using shift software with emission image superimposed (yellow color) before (left) and after shifting emission images (right) showing misregistration area (green) between emission and transmission images.

Mean activity in each quadrant is normalized to the maximum 2% of pixels in the whole heart dataset. Regions of each quadrant are identified having values outside 97.5% confidence intervals (CI) or 2.5 SDs outside normal values of 18 healthy volunteers with no risk factors by complete medical history (no hypercholesterolemia, hypertension, diabetes, smoking, obesity, drug or alcohol abuse, other systemic diseases, or family history of CAD) and the percentage circumferential profile units outside 97.5% CI calculated automatically.

Misregistration Artifacts

Misregistration artifacts were defined qualitatively and quantitatively as (a) abnormalities or defects on perfusion images that did not correspond to any of the coronary artery distribution territories in comparison with our previously reported precise myocardial perfusion map (4); (b) the area of the abnormality on the PET image that was associated with a corresponding area of misregistration of the transmission and emission scans when superimposed; and (c) the defects disappeared and the image normalized when the emission images were shifted to superimpose them precisely on the attenuation images and reconstructed again using the shifted emission data. Healthy control subjects were asymptomatic volunteers with normal rest and stress perfusion images and no risk factors for CAD.

Image Analysis Protocols

To determine the causes or mechanisms of misregistration producing image artifacts, 3 different attenuation-rest-stress imaging protocols were studied.

Quantitative Measurements of Misregistration on PET Scans

The following objective quantitative measurements were made: (i) heart size; (ii) vertical and lateral heart displacement at rest compared with dipyridamole stress; (iii) diaphragm displacement between the initial attenuation scans and between rest and dipyridamole; (iv) size of misregistration of the attenuation and emission images expressed as the percentage of the emission heart area outside the heart contour superimposed on the attenuation image in the horizontal plane; and (v) amount of shifting of the emission image in centimeters that was required to correctly superimpose the attenuation and emission images sufficient to make the artifactual defect disappear (normalize) in the area of prior misregistration on repeated reconstruction using shifted emission data.

These endpoints are listed in a numbered sequence for conceptual clarity. However, for efficiency in image processing and making the measurements, these objective quantitative measurements were mechanically made in detail in a different lettered sequence as follows:

1. Heart displacement in protocols 2 and 3 between rest and stress images was measured by identifying the scanner’s in-plane acquisition tomographic view before rotation showing the uppermost superior border of the heart where each plane has a 5.125-mm slice thickness and a 5.125-mm center-to-center separation.

2. Left diaphragm dome displacement between the 3 initial attenuation scans in protocol 1 and between rest, stress, and late MAC in protocols 2 and 3 was measured by placing the cursor at the top of the right diaphragmatic dome position in the 3-plane display (scanner’s vertical in-plane, vertical, and horizontal longitudinal cuts) of the acquisition tomographic views, before rotation with readout of the cursor position in the x-, y-, and z-planes. Diaphragm motion could not be determined for protocol 2 (early MAC) in the absence of successive rest and stress SAC attenuation images.

3. Quantifying the misregistration area between attenuation and rest and stress emission images was done for protocols 2 and 3 in the 3-plane display of the acquisition tomographic view as follows. Rotation angles were copied from the rest and stress images onto the attenuation image and the attenuation image was rotated exactly the same as the emission images. Emission rest and stress images were thereby superimposed over the attenuation image as originally acquired in exactly the same orientation. To more tightly define the borders of the attenuation image, a 50% cutoff threshold was used for all attenuation images compared with a 30% cutoff threshold for the emission images. Misregistration was defined as the area of the heart on rest or stress emission images that was outside the heart contour on the attenuation image (Figs. 1B, 2B, 3B, and 4B). The misregistration area and the whole heart area were manually outlined with a region-of-interest program that automatically measured the absolute pixel counts of each area. The ratio between the misregistration area and the whole heart area was calculated and expressed as the percentage of the emission heart image outside the attenuation heart image. The whole heart area and misregistration areas were measured twice by the same operator with an intraobserver variability of ≤250 pixels. Misregistration areas of <250 pixels were not considered significant. The heart and misregistration areas were expressed in square centimeters calculated from the reconstructed image pixel resolution of 1.7 × 1.7 mm.

4. Images with misregistration artifacts were displayed in the acquisition tomographic in-plane views using the shift software (Figs. 1B, 2B, 3B, and 4B). The emission images (either rest or stress) were manually shifted in the x- and y-coordinates of the in-plane acquisition view over the attenuation image to achieve visually optimal superposition and proper coregistration of the free lateral heart wall.

5. Completely automated analysis of the artifactual image defect size and severity as previously reported (2–8) and briefly described here was obtained before and after manual shifting of the emission images in the shift view and before correction of the misregistration by repeated image reconstruction.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2.

Severe stress perfusion defect. Legend is the same as Figure 1.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3.

Moderate rest perfusion defect. Legend is the same as Figure 1.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4.

Mild stress perfusion defect. Legend is the same as Figure 1.

Severity of an image defect was quantified as the lowest quadrant average relative activity—that is, average relative activity for the quadrant having the lowest average activity of anterior, septal, lateral, and inferior quadrants. For all patients, the most common, typical, and severe misregistration artifacts were observed in the anterolateral or lateral area of the left ventricle.

Combined size and severity of perfusion defects were defined as the percentage of cardiac image with relative activity of <60% of maximum activity that is 3 SDs below the mean maximum activity of healthy control subjects. The apical quadrant was not included. This measure of perfusion defect severity has been a highly reproducible endpoint with a narrow SD that sensitively quantifies perfusion defects and changes among sequential PET scans with high statistical significance (2–8).

To determine the mechanisms underlying the attenuation and emission misregistration artifacts, a detailed analysis was performed in a subset of 54 patients from the early MAC protocol and in a subset of 55 patients from the late MAC protocol, all with characteristic, definite, and comparable misregistration artifacts. A group of 18 healthy volunteer control subjects imaged using the late MAC protocol was also analyzed. Since protocol 3 (late MAC) had rest SAC, stress SAC, and late MAC attenuation images, a more complete analysis of the mechanisms underlying misregistration of emission images and attenuation was obtained in this group. Additionally, precise measurement of diaphragm motion was obtained using protocol 1 in a group consisting of 45 randomly selected patients, divided in 3 equal groups based on their body mass index (BMI).

Statistical Analysis

All statistical analyses were performed using SPSS version 11.5 (SPSS Inc.). Data are reported as mean ± 1 SD or SEM. Differences among the means of continuous variables for patients and the healthy control group were analyzed with an independent samples, 2-tailed t test using the Levene test for equality of variances. ANOVA was performed for significance of variance with the Games–Howell post hoc test for unequal variances. Pearson bivariate correlation coefficients were calculated. Single stepwise multivariate linear regression analysis was performed to identify predictors for the quantitative degree of misregistration. Multivariate logistic regression analysis was performed to identify predictors for the presence or absence of misregistration artifacts in nonshifted images. A 2-tailed P value of <0.05 was considered statistically significant.

Mechanisms Underlying Attenuation–Emission Misregistration

Our data can be explained by the following synthesis. Systolic contraction causes the root of the aorta and base of the heart to descend and the cardiac apex to recoil forward, up and to the left as in Figure 5, striking the inside of the chest wall where it is felt as an apical impulse. With tachycardia, the heart is in this recoil position for a longer time than that at resting heart rates. Therefore, during the interval of data acquisition for the dipyridamole PET image in which heart rate increases, the heart has a different average position during tachycardia as compared with the normal resting heart rate.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 5.

Normal cardiac and diaphragmatic motion. Black lines show left heart border during diastole and mediastinal outline during expiration. Red lines show changes in left heart border during systole and mediastinal outline during inspiration.

While standing, the diaphragm is at its lowest position and rises toward the heart on assuming the supine position. This upward repositioning of the diaphragm on assuming the supine position may be slow and plastic due to slow-moving abdominal contents toward the heart after lying supine, particularly with the weight of excess visceral fat, as suggested by our analysis of the consecutive initial resting attenuation scans in protocol 1.

Dipyridamole typically causes a feeling of chest fullness and labored breathing associated with tachypnea so that the diaphragm is in a lower position for a longer time than that during normal slow breathing. As the diaphragm shifts inferiorly after dipyridamole, the mediastinum also narrows as illustrated in Figure 6E and Figure 7B.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 6.

Mechanisms for attenuation artifacts using resting attenuation scan. Red lines show positions of heart, mediastinum, and diaphragm during data acquisition; dashed red lines indicate mediastinum and diaphragm during standing and on first assuming supine position (left red dashed lines) and later in supine position after abdominal content shift cephalad (right red dashed lines). Black lines show positions of those structures not being imaged during attenuation or emission data acquisition of structures in red. Attn = attenuation; Dipy = dipyridamole.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 7.

Mechanisms for attenuation artifacts using postdipyridamole scan. Legend is the same as Figure 6.

Therefore, the attenuating structures of the chest—lungs, left mediastinal border, heart, and diaphragm—undergo substantial slow plastic changes on assuming the supine position during resting and during dipyridamole imaging. Since the attenuation and emission data are acquired at different times during this complex sequence of slow plastic changes in the attenuation structures, attenuation–emission misregistration may occur (Figs. 6 and 7).

Using the early resting measured attenuation data for the dipyridamole emission image may cause misregistration as illustrated in Figure 6. Using the late measured attenuation after dipyridamole for the dipyridamole emission scan reduces misregistration artifacts in stress emission images since the attenuation structures best approximate those during stress emission image acquisition. However, using the late attenuation after dipyridamole for the resting emission scan may then cause attenuation–emission misregistration and artifactual defects in the rest emission scan, as illustrated in Figure 7.

The correlation between misregistration and BMI may be explained by several potential mechanisms. Greater BMI indicates more visceral fat having greater effects on the diaphragm position associated with greater attenuation–emission misregistration and more artifactual defects. In addition to these complex position changes, the amount of fat around the heart also has significant effects on attenuation–emission misregistration. In obese people, the fat around the heart causes the borders of the attenuation image to be bigger than the borders of the heart. Therefore, the left mediastinal border of the attenuation image extends farther leftward than the myocardium and thereby produces a “halo” of attenuation-corrected space around the heart that further complicates attenuation correction. The association of small hearts with more attenuation–emission misregistration is likely due to the relatively poor resolution of PET compared with small hearts such that even slight absolute misregistration causes a relatively large percentage of the heart to be misregistered.

Lateral cardiac recoil and diaphragmatic displacement have different effects on attenuation–emission superposition depending on whether the heart is positioned vertically as in thinner subjects or horizontally as in heavier people. For a horizontal heart as occurs in heavier people, recoil and diaphragm displacement tend to cause vertical misregistration with an anterolateral artifact, whereas in a vertical heart the artifact is more lateral. In thin people, misregistration artifacts are more likely due to cardiac recoil and downward diaphragmatic displacement after dipyridamole. In heavier people, these artifacts are more likely due to slow plastic upward displacement of the diaphragm and visceral fat on assuming the supine position plus later the same above changes after dipyridamole.

z-Axis misaligment along the long axis of the scanner is important as our illustration shows. However, as illustrated in Figures 5–7, the long axis and the left border of the heart are slanted leftward and downward so that horizontal shifting of the images in the in-plane acquisition tomographic views also corrects for vertical misalignment of attenuation and emission images as our data demonstrate. z-Axis shifting along the long axis of the scanner in 3D PET acquisition without in-plane septa is easier than in 2D PET acquisition with in-plane septa; however, the scatter from other organs in 3D PET is a major problem that substantially degrades cardiac images in our experience.

Solutions for Attenuation–Emission Misregistration

These complex changes in the attenuation structures of the thorax preclude predicting attenuation–emission misregistration artifacts in any individual. However, an attenuation–emission imaging sequence can be chosen to reduce misregistration artifacts that can also be recognized, routinely assessed on clinical images, and corrected. In our experience, the sequence of a single initial resting attenuation scan followed by resting and dipyridamole emission scans has the highest probability of attenuation–emission misregistration artifacts. For a one-attenuation scan protocol, the sequence of rest and dipyridamole emission scans followed by a late measured attenuation scan starting approximately 5–10 min after the dipyridamole emission scan has fewer misregistration artifacts since the late attenuation scan better approximates the average attenuation structures over the period of emission imaging.

Other solutions include separate measured attenuation scans for rest and for dipyridamole as we now currently do for more reliable results despite slightly prolonging the study. However, we also check every image with any defect for correct attenuation–emission coregistration using the acquisition tomographic shift software and shift the emission images to achieve optimal coregistration and reconstruct the emission images with shifted data if needed.

In our experience a short SAC attenuation scan immediately after dipyridamole frequently causes attenuation–emission misregistration for several complex interacting reasons. The attenuation structures during this brief SAC acquisition do not approximate the average of the changing attenuation structures during emission scanning or we do not adequately understand the SAC method yet. Another approach to attenuation correction in PET may combine CT and PET scanning that also reduces the study time (23–25). However, the same potential problems may be exaggerated by the very brief CT attenuation acquisition that does not approximate the average of the dynamically changing attenuation structures present during the more prolonged emission data collection. Further studies of the combined PET/CT approach are needed.

Limitations of Study

An expected criticism of this study is lack of coronary arteriography to prove that the image abnormalities were artifactual and are not due to CAD. Our study has demonstrated artifactual defects due to attenuation–emission misregistration on resting images that improve or disappear on dipyridamole images in which the attenuation–emission registration is corrected. The disappearance of resting image abnormalities on stress images after dipyridamole is not consistent with flow-limiting coronary artery stenosis as justification for coronary arteriography. For assessing the less severe perfusion abnormalities seen in early preclinical coronary atherosclerosis, coronary arteriography cannot be justified. Furthermore, based on >1,000 patients with coronary arteriograms, we have previously published the diagnostic value of mild-to-moderate regional PET perfusion defects and of the base-to-apex longitudinal perfusion gradient associated with diffuse nonobstructive coronary atherosclerosis before significant localized stenosis develops (1–3). By demonstrating specific attenuation–emission misregistration in the same region as the abnormality on the PET images and normalization of the images when the misregistration is corrected, we believe that our results do not require coronary arteriographic confirmation. Moreover, we have shown the longitudinal base-to-apex perfusion gradient to be a significant marker of diffuse CAD before significant flow-limiting coronary artery stenoses occur (3). This observation combined with the well-documented gross inaccuracies of visual arteriographic interpretation could be viewed as PET providing a more reliable reference standard for severity of coronary disease than the coronary arteriogram as used now clinically, provided attenuation–emission images are correctly coregistered.

A second criticism of the study is the absence of complete information on diaphragmatic displacement on assuming the supine position or after dipyridamole in the 3 different imaging protocols reported here. Traditionally, the cardiac PET imaging protocols involved only a single initial resting measured attenuation scan followed by rest and dipyridamole emission scans. For patients studied using this traditional single resting attenuation image before May 2000, we recognized a potential problem but had no knowledge of the causes or solutions. By May 2000, at the beginning of the 1,177 patients in this study, we had discovered the attenuation–emission misregistration problem and its empiric correction by shifting the emission images to obtain coregistration but without understanding the mechanisms. With further experience we have developed multiattenuation emission image sequences to study the mechanisms and solutions as reported here.

As a consequence of this study, we now routinely use a protocol with the following image acquisition sequence: a resting measured attenuation, resting emission, dipyridamole emission, and a postdipyridamole measured attenuation. Our study demonstrates a concept where the details of thresholding, viewing, and coregistering emission and attenuation images may vary depending on the scanner, the radionuclide used, the reconstruction technique, resolution settings, and the goal of cardiac PET in each case.