Effect of Acquisition Orbit on SPECT in Phantoms

TO THE EDITOR:

The recent article by Liu et al. (1) presents data that challenge widely accepted standards of practice used in most clinical rotational SPECT myocardial perfusion protocols performed today. The focus of that article was on the increased inhomogeneity of circumferential count profiles (CCP) acquired with a 180° orbit in comparison with profiles from studies acquired with a 360° orbit. This observation was based on simplistic phantom data acquired using 1 specific dual-head SPECT camera (Millennium VG; General Electric Medical Systems). The authors also mentioned correlative data acquired using a triple-head camera as being “entirely consistent with those acquired by the dual-head system,” and if we accept their conclusion without qualification, the current standard of clinical practice should immediately be revised to adopt 360° orbits in deference to 180° orbits.

We are particularly concerned that the CCP curves from the reconstructed phantom data taken in a “centered” position (their Fig. 4) demonstrated a 3-cycle sinusoidal “wavy” artifact (WA) averaging about 15% peak-to-peak amplitude, which the authors described as “presumably caused by photon scatter and self-attenuation and perhaps by imprecise positioning of the phantom.” They went on to include the observation that the “wavy ′normal’ circumferential count profiles of the phantom in the center position may be caused by a slight deviation of the long axis of the phantom from the centerline of the table,” and as further explanation for this WA, they “noted a slight wobble on cine display” observed during playback of the phantom data. The authors were aware that rotational tomographic imaging of this phantom, symmetrically positioned at the center of the orbit, should produce essentially flat or uniform CCP curves; however, this discrepancy in their data did not prompt further investigation. They did note that this variability was considerably worse (30% peak-to-peak) for 180° data collected 15 cm off the center of rotation. This result was consistent with distance and resolution variations occurring throughout the off-center orbit. These variations tended to balance out for the 360° orbit, but the WA is still present, although to a lesser degree.

Among the significant impediments facing contemporary nuclear cardiology is the unimpressive specificity and reproducibility consistently demonstrated in the diagnostic accuracy reported for clinical myocardial perfusion studies. We have observed that many definitive studies, including that of Liu et al. (1), pass by the opportunity to conduct further objective testing that might shed important light on some of the fundamental shortcomings of rotational SPECT. Another recent article with which we take exception, by Blagosklonov et al. (2), stopped just short of running correlative tests and performing additional analytic steps, which could be very helpful in enhancing our understanding of the root causes contributing to the disappointing specificity and reproducibility experienced in the clinical application of myocardial perfusion studies.

We feel strongly that the WAs noted in this article (1) are not a minor issue and that their presence casts doubt on the validity of the data and the conclusions presented there. To expand on this point regarding the data presented in this article, it is important to identify the source of the WA as originating from 1 of 3 nominal error sources (systematic, experimental, or analytic). One fact that is clear from the article concerns the angular dependence of the problem, and so one should focus primarily on investigating those aspects of system performance, experimental design, and data analysis that can potentially alter resolution, sensitivity, or geometric registration as a function of viewing angle. Within this framework, we suggest 3 areas for further investigation that might be helpful in at least identifying the level of the problem: systematic errors, experimental errors, and analytic errors.

Regarding systematic errors, those aspects of system performance that are most likely to exhibit angular dependence include problems related to (a) the precision and uniformity achieved in the construction of the collimators, (b) the electromechanical precision, the pointing accuracy and reproducibility with which the detectors are moved through their orbits, and (c) the electrical stability of the images registered in the computer from the detectors. These types of problems can usually be selectively investigated by imaging point sources in appropriate configurations—for instance, (a) imaging a point source at a distance of 3–4 m to check collimator performance, (b) at the center of rotation to check the pointing accuracy and reproducibility of the orbit, and (c) fixed to the surface of the collimator to check image registration. Cine playback of the point source data provides a quick-look option for demonstrating obvious problems, but detection of more subtle sources of error requires customized software to perform rigorous analysis of peak location, center of mass, full width at half maximum, and the integral sensitivity of the individual images. Note that support stands which hold the detectors in a cantilevered position, such as the SPECT system used in the study of Liu et al. (1), are more susceptible to errors in the pointing accuracy associated with the collected views.

Regarding experimental errors, certain aspects of the experimental setup used in the work of Liu et al. (1) need to be looked at more closely, including the test fixture (clamp) used to hold the phantom, which appears to extend longitudinally some distance along the base (open end) of the phantom and places attenuating and scattering material (compressible layers and wood clamps) close to the radioactive wall of the phantom. This is of particular concern because the configuration of the clamp has cyclic rotational symmetry, which conceivably could cause a high-order, multiple-scatter phenomenon resulting in the nonuniform pattern seen in their Figures 4 and 5. This possibility could be excluded by replacing their clamp with a foam or fabric cradle suspended between 2 thin rods to facilitate acquisition of a set of images with minimal scatter and attenuation.

Another experimental issue that needs attention is the use of a 15% symmetric energy window. The digital correction methods currently applied to the energy signal from state-of-the-art detectors allow much greater flexibility in setting energy windows than was formerly possible. Given that no other scatter reduction techniques are applied, setting an energy window that is asymmetrically high relative to the photopeak has been shown to improve the primary-to-scatter ratio and, therefore, resolution. The presence of angular dependence in the scatter ratio opens the possibility that use of a narrow, symmetric energy window could result in angular dependence of the system resolution and, thereby, nonuniformity in the reconstructed images and the CCP curves.

Regarding analytic errors, among a wide array of possibilities are some subtle pitfalls associated with CCP curves. Our own experience (3) has shown that artifactual nonuniformity can occur when CCP curves are developed by plotting the single value of the maximum pixel found along each individual search radius. Under certain conditions depending on the relationship between the exact physical position of the phantom in 3-dimensional space and the centers of the voxels in the reconstruction data matrix, this alignment factor can result in significant cyclic variations (essentially a Moiré effect) in the values of the maximum pixels extracted for the CCP curves and provides another possible explanation for the nonuniformity in the CCP curves presented by Liu et al. (1). One simple solution is to plot the maximum plus the sum of its 2 nearest radial neighbors, introducing a smoothing effect that will minimize the WA. The work of Liu et al. used the WL-CQ program (Eclipse Systems), which was described as using “maximal pixel values” to develop the CCP.

It is important to realize that this type of simplistic, thin-walled phantom is a very rigorous testing device and we should be careful about extrapolating any test results obtained with it to have immediate clinical implications. Every commercial SPECT system in clinical use today introduces its own particular forms of “distortion” in application to the acquisition, reconstruction, and presentation of 3-dimensional images. However, these systems can still have utility and efficacy in clinical application so long as the users are aware of these characteristics and are willing to develop and maintain techniques to recognize and “read around” these artifacts (4). Some of the measures that have been proposed to minimize artifacts (attenuation correction, elliptic orbits, and prone imaging, to mention a few) have not been proven to consistently increase the accuracy and reliability of SPECT cardiac studies.

Taken at face value, the single set of patient images shown in the article of Liu et al. (1) (their Fig. 1) implies that false-positive clinical results can primarily be attributed to the use of 180° orbits in performing myocardial perfusion studies. We cannot objectively evaluate this conclusion, however, until the WAs in the CCP curves shown there are fully explained. Other clinical investigations have shown that equally deleterious effects result from the poor resolution and scatter environments encountered under nominal clinical imaging conditions, especially when imaging large patients using a 360° orbit. What concerns us more than the issue of orbits is the ongoing failure to accurately identify all the fundamental problems associated with rotational SPECT contributing to the unacceptably high incidence of false-positive studies.

The field of nuclear cardiology has experienced significant frustration and more than a few credibility issues stemming from the inherent variability in the anatomy and physiology of the human cardiovascular system as it interacts with the characteristics of rotational SPECT imaging systems to create artifacts. We have similar concerns with regard to the article by Blagosklonov et al. (2), describing “motionlike artifacts” in rotational SPECT studies. We disagree with the conclusion of those authors and propose that poststress, positionally dependent cardiac volume changes cause inconsistencies in the SPECT data that are much more problematic in the manner in which they temporally and spatially interact with sequential rotational SPECT image acquisition to create artifacts.

Rather than the orbital issues discussed in the article by Liu et al. (1) or rapid early ²⁰¹Tl washout as postulated in the article by Blagosklonov et al. (2), cardiac volume variability, which occurs after stress and after an acute transition from an upright to a supine position, is a more likely explanation for the compromised specificity and reproducibility encountered in rotational SPECT imaging. With regard to the clinical images shown in Figure 1 of Liu et al. (1), the 360° data were most likely taken over a longer interval than were the 180° data and therefore were less subject to the acute changes in cardiac volume that might have occurred immediately after the healthy volunteer was placed supine. This could be an alternative explanation for the artifact observed in the 180° data of their Figure 1.

Keep in mind that myocardial perfusion imaging primarily is a diagnostic modality that uses individual patients as their own standard. In other words, we are interested in comparing patients studied under stress conditions with the same patients studied at rest. The most salient fact apparent from the article of Liu et al. (1) is the technical importance of performing reproducible stress/rest imaging sequences on the same patient, using the same patient location and orientation and the same camera orbit for acquisition of both datasets, thereby minimizing systematically induced variables. The point that needs to be emphasized is that, whatever the artifactual characteristics of a given imaging system, the clinical efficacy of that system will largely depend on the user’s ability to reproduce or eliminate those artifacts in successive image-acquisition sessions, thereby ensuring diagnostic comparability and accuracy. Far more significant advantages would be gained by such simple and expedient measures as marking patients with a laser positioning device to ensure reproducible positioning, using breast binding on female patients, and achieving physiologic equilibrium by having patients lie down for 15 min before initiation of rotational SPECT acquisitions.

The field of nuclear cardiology is currently at an impasse because it is relegated almost exclusively to the use of rotational SPECT imaging devices, which are dependent on moving detector systems for imaging the heart. The heart is itself a dynamically changing object in terms of both its size and its position. To a limited extent, temporal changes in patient position are correctable with software but changes in the size of the heart with time probably are not. Under these circumstances, one would expect some fundamental limitations to the accuracy and reproducibility of the results achievable by any system that acquires image data sequentially in time and immediately after the patient lies down. Whatever imaging methodology and clinical protocol is used, however, requisite technical measures must be taken to maximize statistical accuracy and ensure adequate image reproducibility.

One feasible design approach in this application is embodied in a multipinhole system based on arrays of stationary detectors that produce simultaneous stress/rest images within a single data-acquisition session (3). This type of simultaneous imaging system eliminates inconsistencies within the image dataset caused by temporal changes that occur between images or imaging sessions. The work of Liu et al. (1) has indicated that trade-offs between resolution, cost, and performance can result in a functional design for a multidetector, multiview system that does not need to acquire data from a full 360° subtended angle surrounding the patient. A primary benefit of this approach is an order-of-magnitude improvement in the statistical significance and quality of the gated images produced, and a direct result of the improvement in statistics is the ability to perform simultaneous stress/rest ²⁰¹Tl/^99mTc perfusion imaging (5). Our experience is that this is the key to improved clinical specificity and reproducibility, which could open the door to the next level in the future practice of nuclear cardiology.