Simultaneous Transmission/Emission Myocardial Perfusion Tomography
Diagnostic Accuracy of Attenuation-Corrected 99mTc-Sestamibi Single-Photon Emission Computed Tomography
Background The purpose of the present study was to assess the diagnostic performance of attenuation-corrected (AC) stress 99mTc-sestamibi cardiac single-photon emission computed tomography (SPECT) for the identification of coronary heart disease (CHD).
Methods and Results With a triple-detector SPECT system with a 241Am transmission line source, simultaneous transmission/emission tomography (TCT/ECT) was performed on 60 patients with angiographic coronary disease and 59 patients with ≤5% likelihood of CHD. Iteratively reconstructed AC stress 99mTc-sestamibi perfusion images were compared with uncorrected (NC) filtered-backprojection images. Normal database polar maps were constructed from AC and NC images for quantitative analyses. From the low-likelihood patients, the visual and quantitative normalcy rates increased from 0.88 and 0.76 for NC to 0.98 and 0.95 for AC (P<.05). For the detection of CHD, the receiver operating characteristic curves for the AC images demonstrated improved discrimination capacity (P<.05), and sensitivity/specificity values increased from 0.78/0.46 (NC) to 0.84/0.82 (AC) with visual analysis and from 0.84/0.46 (NC) to 0.88/0.82 (AC) with quantitative analysis. For localization of stenosed vessels, visual and quantitative sensitivity values were 0.51 and 0.63 for NC and 0.64 and 0.78 for AC images (P<.05), respectively.
Conclusions TCT/ECT myocardial perfusion imaging significantly improves the diagnostic accuracy of cardiac SPECT for the detection and localization of CHD. Clinical use of TCT/ECT imaging deserves serious consideration.
Myocardial stress perfusion imaging is widely used to noninvasively detect the presence and extent of CHD. Although perfusion imaging with SPECT has been shown to be highly sensitive in the detection of CHD, its specificity in assessing the absence of disease has been suboptimal.1 2 3 It has been hypothesized that the suboptimal specificity of myocardial SPECT perfusion imaging is due to the inhomogeneous attenuation of photons in the thorax, which produces artifactual heterogeneous radionuclide activity distributions in the myocardium of patients without evidence of CHD. Although guidelines have been adopted to distinguish attenuation artifacts from real perfusion abnormalities,4 5 6 7 photon attenuation reduces the discriminating capacity of myocardial perfusion SPECT, resulting in lowered test specificity.
Recent research efforts have resulted in the commercial availability of SPECT imaging systems that are capable of TCT.8 9 10 11 Whether performed sequentially or simultaneously, TCT provides anatomically specific density maps of the thorax that can be used to correct SPECT image data for photon attenuation. We developed a system that uses a triple-detector system with a 241Am transmission line source (activity, 230 MBq/cm) opposite a fan-beam collimator (focal length, 65 cm) on one detector to collect transmission and emission projection data.12 13 An Americium-241 line source was chosen based on its low photon energy (60 keV), which permits its use with both 201Tl and 99mTc radiotracers with minimal cross-talk contamination of the emission data. Parallel-hole collimators instead of fan-beam collimators were used on the “emission-only” detectors to minimize the truncation of emission data caused by the inherent magnification of fan-beam collimators and the image artifacts that can result.
The purpose of the present study was to assess the diagnostic performance of this TCT/ECT imaging system for the detection and localization of CHD in patients with angiographically documented CHD and in patients with a very low likelihood of CHD. NC and AC emission tomograms were visually and quantitatively interpreted and compared for the presence or absence of CHD by using ROC curve analysis.14 ROC curve analysis was used to characterize and compare the inherent discriminating capacity of the NC and AC images. Normal NC and AC 99mTc-sestamibi perfusion distribution patterns from a database of patients with a very low likelihood of CHD were used in the quantitative analysis of patients with suspected CHD and to determine normalcy rates in low-likelihood patients. Based on ROC curve analyses from the quantitative interpretation of perfusion images, we determined optimal abnormality thresholds for the AC and NC images.
Simultaneous TCT/ECT was performed on a random selection of 360 patients who were referred for stress myocardial perfusion imaging. From this population, two groups of patients were extracted for this study. Group 1 was composed of 59 consecutive patients (22 men and 37 women; mean age, 53±14 years) who had ≤5% pretest likelihood for CHD15 (mean±SD, 2.0±1.5%). For quantitative analysis, sex-specific normal databases of 20 male and 20 female patients were constructed from AC and NC images. Normalcy rates were determined for group 1 patients. Group 2 was composed of 60 consecutive patients (38 men and 22 women; mean age, 63±12 years) who had angiographic verification of disease within 90 days of scintigraphy without any change in cardiac condition or treatment. The mean time interval between tomography and catheterization was 15±20 days. Within this group, 25 patients (17 men and 8 women; mean age, 60±16 years) had historical and ECG evidence of a prior myocardial infarction. From the random patient population (n=360), no patients were excluded from either group 1 or group 2 due to image quality, patient size, or location and extent of perfusion defects, and no patient was excluded because of coexisting hypertensive heart disease, ventricular dysfunction, or dilation.
Patients received 111 MBq (3 mCi) thallous chloride for the resting perfusion study. The stress imaging commenced immediately after rest or redistribution imaging. At peak stress, 1110 MBq 99mTc-sestamibi (DuPont Pharma) was injected. Pharmacological stress testing was performed with adenosine16 (n=61), dipyridamole17 (n=12), or dobutamine18 (n=7). Symptom-limited maximal treadmill exercise was performed using the Bruce protocol (n=39).
Coronary angiography was performed using standard percutaneous techniques. Coronary stenoses were assessed by two experienced angiographers and expressed as percent luminal diameter stenosis; the angiographers were unaware of the scintigraphic results. From the angiographic data, coronary stenosis severity was scored using the following system: 1, 0% to 49%; 2, 50% to 69%; 3, 70% to 89%; and 4, 90% to 100% reduction in luminal diameter. For comparison with the scintigrams, left main coronary lesions were recorded as both LAD and LCx stenoses. Diagonal branches were included in the LAD, and marginal branches were grouped with the LCx. Lesions to the posterior lateral and posterior descending arteries were included with the RCA. Perfusion SPECT data were analyzed at angiographic thresholds of ≥50% and ≥70% stenoses of luminal diameter.
Acquisition and Processing of TCT/ECT Data
Each patient underwent simultaneous TCT/ECT imaging with a collimated line source positioned opposite and parallel to one of three detectors on either a PRISM 3000XP or a PRISM 3000 SPECT imaging system (Picker-Ohio Imaging, Nuclear Medicine Division of Picker International). This detector was fitted with a low-energy, high-resolution fan-beam collimator with a focal length of 65 cm. The transmission source consisted of a 5.55-GBq (150 mCi) Americium-241 line source (Isotope Products Laboratories) sealed in a 0.3-mm-thick stainless steel tube with an active volume of 2.4 mm (ID)×240 mm. The motor-driven line source holder provided radial and axial collimation while enabling noncircular orbit acquisitions. The remaining two detectors were fitted with low-energy, high-resolution, parallel-hole collimators to collect uncontaminated, untruncated emission data.
Transmission and emission projection data were stored in 64×64 matrices for all detectors. Projection images were acquired at 6° steps over a 360° orbit for 12 seconds per projection for a total scan time of 12 minutes. To improve transmission counting statistics, patients weighing more than 90 kg were imaged for 16 seconds per step. An energy window of 59.0±5.9 keV was used for the 241Am transmission images. Typical transmission count rates varied from 6 to 30 1000 counts/s (kcps) depending on patient size, positioning, and density. For the 99mTc-sestamibi stress study, an energy window of 140±10.5 keV was set to collect emission data. The projection data from the three detectors were acquired simultaneously in both the transmission and emission windows. A 20-minute transmission “blank” scan (≈30 million counts acquired once per week) was acquired to compute attenuation line length sinograms from the transmission data. To minimize potential dead time effects during the blank scan, the source was shielded to reduce the beam intensity to less than 60 kcps.
After removal of cross-talk contamination from the transmission and emission data, attenuation maps were reconstructed from the truncated transmission data using a penalized, weighted, least-squares algorithm.19 20 The boundary constraint for the reconstruction was the patient outline and the known position of the imaging table. The patient outline was estimated from the emission projection data. The initial estimate to the algorithm was constructed from a FBP reconstruction of the transmission data with the truncation ring artifact removed and the area between the FBP image and patient outline assigned the attenuation coefficient of tissue. Before reconstruction of the emission data, the uncontaminated emission projection data from the parallel collimated detectors were summed. The fan-beam emission data were not included in the reconstruction to minimize possible truncation artifacts in large patients whose fan-beam emission data would be severely truncated. AC emission images were constructed using a penalized, weighted, least-squares algorithm with the reconstructed attenuation maps to correct the emission data for photon attenuation. The initial estimate for the reconstruction of the AC emission images was a first-order Chang-corrected FBP image.21 From this initial estimate, transaxial emission images were reconstructed in 12 iterations. The average processing time for 18 AC emission images was ≈2.5 minutes.
NC emission images were reconstructed using conventional FBP with a ramp filter having a cutoff equal to the Nyquist frequency fn. Negative image pixels were not zeroed. Before reslicing, the NC and AC images were filtered with a three-dimensional Butterworth filter of order 5 with a frequency cutoff of 0.6∗fn. NC and AC emission images were reformatted into short-axis and horizontal and vertical long-axis slices for visual and quantitative analyses. To normalize the resliced images for visual display, voxel counts outside of the heart were truncated to the maximum pixel count within the myocardium.
All NC and AC images were interpreted by two observers who were blinded to the angiographic results. The resliced image sets from group 1 and 2 patients were pooled and randomly presented to the observer. The NC and AC image sets were presented to the observers on separate days. While interpreting the images, the observers were aware of patient sex and the type of reconstruction but were unaware of the interpretation of the paired study (AC and NC). In scoring the images, the basal, mid, and distal short-axis slices were divided into six evenly spaced segments, and a single segment representing the apex was scored from the long-axis sections. Tracer uptake in each sector was scored as 1=normal, 2=equivocal, 3=abnormal, or 4=severely abnormal or absent. Each vascular territory was assigned the maximum score of the segments within its range. Territories with discordant readings were reread and assigned a consensus value. A vascular territory was considered abnormal if one of the sectors within its boundaries received a score ≥2. Likewise, the study was considered abnormal if any vascular territory was scored as abnormal.
Quantitative results were generated from the tracer distributions in polar map format compared with a database of normal distributions.1 22 23 The distal-to-basal rings of the polar maps were constructed from maximum-count circumferential profiles of short-axis slices. The apex was similarly sampled from vertical long-axis slices. The polar maps were normalized to the maximum count within each map. To account for differences in the photon attenuation patterns of male and female patients, gender-specific databases (n=20) were constructed for the quantitative analysis of NC images. Gender-specific (n=20) and gender-composite normal databases (10 men and 10 women) were constructed for the AC analyses. Polar map pixels ≥2.5 SD below the normal mean of the appropriate database values were scored as abnormal. The vascular territories used to determine the localization and extent of the perfusion defects in these maps were defined as described by Maddahi et al.23 ROC analysis was used to determine the abnormal extent thresholds in each vascular territory of the AC and NC polar maps.
All data are expressed as mean±SD. The normalcy rate as determined from group 1 patients was defined as the ratio of the number of patients with normal tomograms to the total number of patients in the group. Data from group 2 patients were used to construct ROC curves and to calculate sensitivity, specificity, and accuracy values. Sensitivity was defined as the number of true-positive tomograms divided by the number of patients with angiographic disease. Specificity was defined as the number of true-negative tomograms divided by the number of patients without disease. ROC curves were constructed by plotting the true-positive fraction (TPF=sensitivity) versus the false-positive fraction (FPF=1−specificity) for different abnormal thresholds. The χ2 test was used to determine the significance of differences in the normal coronary polar map distributions. McNemar’s test24 was used to determine differences in paired data samples. The corroc program25 was used to fit the ROC curves, calculate the integral areas of the curves, and estimate the significance of differences between AC and NC curves.
Normal Tracer Distributions
Gender-specific databases of normal myocardial tracer distributions were computed and displayed in polar map format (Fig 1⇓) for the AC and NC images. Differences between the NC patterns of the two sexes were most apparent in the lateral and posterior walls of the myocardium, where the normalized male distribution was 7% to 9% lower (P<.001) than the normalized female distribution due to soft tissue attenuation. Because there were no significant differences between the male and female AC distributions, a gender-composite distribution was constructed. The AC normal polar maps have improved homogeneity in the basal and distal rings compared with the NC polar maps. The anterior-to-posterior activity ratio in basal short-axis sections decreased from 1.34±0.11 for the normal male NC distribution to 1.00±0.08 for the AC distribution (P<.001). Compensation for basal attenuation was also demonstrated by a decrease in the distal-to-basal activity gradient in the posterior wall from 1.15±0.05 for the NC images to 1.03±0.05 for the AC images (P<.001). The 10% to 15% reduction in tracer activity at the apex of the AC distribution is most likely due to the normal anatomic thinning in this area and the partial volume effect that is inherent to SPECT imaging.26
The angiographic findings for the group 2 patients are summarized in Table 1⇓. Two patients did not have their RCAs injected; therefore, visual and quantitative radionuclide data for these distributions were not included in the analysis. Two patients had dominant left coronary arteries, so the territory commonly associated with the RCA was assigned to the LCx. Four vascular territories in patients whose vessels were unstenosed and had had previous myocardial infarction were omitted from the analysis. For patients with significant stenoses, the mean number of diseased vessels was 2.3±0.8 for a stenosis cutoff of ≥50% and 2.1±0.8 for a stenosis cutoff of ≥70%. The mean severity score for patients with disease was 2.8±0.8. For each coronary distribution, the mean severity scores were 2.7±1.2 for the LAD, 2.6±1.2 for the LCx, and 3.2±1.2 for the RCA. Differences in severity scores were not significant.
Visual Analysis of Images
Visual interpretation of the group 1 patients demonstrated a significant increase in the observed normalcy rate (AC, 0.98±0.02; NC, 0.88±0.04; P=.027). In the visual analysis of the NC images, there were seven false-positive readings, two involving the LAD and five involving the RCA. Both of the LAD abnormalities were due to breast attenuation in women. Four of the five false-positive readings in the RCA territory could be attributed to diaphragmatic attenuation; the AC images showed normal activity in each case. NC and AC images from two of these false-positive patients are displayed in Fig 2⇓. The remaining false-positive study from the NC images was also read as abnormal from the interpretation of the AC images (LCx and RCA territories). This patient had a history of dilated cardiomyopathy, which was the most likely cause of the inhomogeneity in tracer distributions (NC and AC images).
Visual analysis was performed using the AC and NC images from the group 2 patients, and ROC curves were plotted for the detection of CHD and for the localization of disease in each of the three main coronary distributions (Fig 3⇓). Curve fits were computed for each set of ROC data points and displayed on the ROC plots. The constant curve fit for the uncorrected LAD data was replaced with hand-drawn curves. The integral areas (Az) for the AC and NC curves were calculated with an estimate of the significance of the difference (P) between the AC and NC Az values.
There was a significant improvement in the discriminating capacity of the AC images compared with the NC images for the detection of CHD, as demonstrated by the increase in the Az value (Fig 3⇑). For the localization of disease, a significant increase in Az was seen in the RCA distribution. This increase is attributed to a decrease in false-positive and false-negative readings resulting from diaphragmatic attenuation. The increased Az value for the LCx distribution was not significant based on the P value and the statistics of the study. Because the NC LAD data were not fit by ROC analysis, the NC Az values were not calculated for comparison of the LAD territory data.
Sensitivity, specificity, and accuracy values are given in Table 2⇓ for the detection of CHD, localization of disease in general, and localization in the LAD, LCx, and RCA territories specifically. For patients with SVD and MVD, sensitivity values are reported for the detection of disease in general in addition to localization of stenosed vessels in these patients. With the exception of the specificity values in the LAD territory, there were increases in each of the variables at both ≥50% and ≥70% stenosis breakpoints. The slight decrease in specificity for the localization of disease in the LAD territory was not statistically significant. Statistically significant differences (P<.05) between AC and NC imaging were apparent in the measured accuracy for the detection of CHD, localization of stenosed vessels in general, and localization of disease in the LCx and RCA territories specifically. Significant improvement with the use of AC images was also demonstrated in localizing diseased vessels in patients with MVD. Specificity, and therefore accuracy, values were not reported for the detection of CHD in patients with SVD and MVD as these patients were a diseased subset of group 2 and specificity values are based on patients without disease.
Quantitative Analysis of Images
From the quantitative analysis of the AC and NC images from the group 2 patients, ROC curves were plotted for the detection of CHD and for the localization of disease in the LAD, LCx, and RCA territories (Fig 4⇓). Curve fits were computed for each set of ROC points and displayed on ROC plots with their corresponding Az and P values. Increases in Az values were seen with AC in all of the observed ROC curve pairs for both angiographic criteria. Statistically significant differences were demonstrated for CHD detection in general at both angiographic cutoffs and for the localization of disease in each of the vascular territories using an angiographic cutoff of ≥50% stenosis. The LCx territory also demonstrated a significant improvement at the ≥70% stenosis cutoff.
From the quantitative AC and NC ROC data, the optimal defect size for the localization of disease in each vascular territory was computed by maximizing diagnostic accuracy. The computed thresholds for AC images were 12% to 24% for the LAD, 12% to 28% for the LCx, and 8% to 12% for the RCA territories. For the NC images, the computed thresholds were 12% to 20% for the LAD, 12% to 28% for the LCx, and 0% to 8% for the RCA. For each of these threshold ranges, the lower limit had higher sensitivity at the expense of lower specificity. From these ranges, defect thresholds of 12% for the LAD, 12% for the LCx, and 8% for the RCA were selected for the analysis of both AC and NC defect maps. These defect thresholds were chosen from the measured ranges for the following reasons: (1) it is better to operate at higher sensitivities with a marginal decrease in specificity because the prevalence of disease in our clinical population is relatively high, (2) these thresholds are identical to the established cutoffs for NC defect maps commonly used,23 permitting comparisons with published data, and (3) the statistical significance of the results presented in the present study were invariant to the chosen defect thresholds (several threshold combinations in the above ranges were tested).
With the above defect size thresholds in the quantitative analysis of the group 1 patients, there was a significant increase (P=.003) in the observed normalcy rate from the interpretation of AC images (0.95±0.03) compared with NC images (0.76±0.06). As seen in the visual analysis, statistically significant increases in diagnostic accuracy were demonstrated using AC images for the detection of CHD, for the localization of disease in general, and for localizing disease in the LCx and RCA vascular territories specifically (Table 3⇓). Quantitative interpretation of AC images also significantly improved the diagnostic accuracy in localizing stenosed vessels in patients with MVD. For patients with SVD, sensitivity and specificity values were slightly higher using NC images compared with AC images, but these increases were not significant. Data in this table also demonstrated that the quantitative interpretation of AC images did not significantly improve the detection of CHD in patients with SVD or MVD.
The simultaneous TCT/ECT imaging system used in the present study demonstrated improved diagnostic accuracy for the identification and localization of CHD. As a result, there were significant improvements in the observed normalcy rates of low-likelihood patients when using AC images compared with NC images. Increases in the observed normalcy rates were evident by both visual and quantitative analyses. With either visual or quantitative analysis of the AC images, there were increases in sensitivity, specificity, and accuracy for the detection and localization of CHD. The most significant increases in accuracy were apparent in the detection and localization of coronary stenoses of intermediate severity (≥50% and ≤70%).
Detection of CHD
Both visual and quantitative ROC curves demonstrated significant improvements in the discriminating capacity of AC imaging compared with NC imaging. This was evident by the larger Az value for AC images for the detection of CHD. Larger Az values demonstrate greater separation between diseased and healthy coronary distributions, resulting in better discrimination between stenosed and unstenosed states. It is also important to note that the improvement of the AC curve compared with the NC curve was greater at the ≥50% stenoses cutoff (+20%, P=.005) in comparison to the improvement at the ≥70% stenoses cutoff (+12%, P=.016). This is indicative of a significant increase in diagnostic accuracy of AC imaging compared with NC imaging, especially for coronary arteries with stenoses in the range of 50% to 70% compared with more severe stenoses.
The characteristics of the AC data produced statistically significant increases in diagnostic accuracy with ROC analysis for the detection of CHD at both angiographic cutoffs investigated. The improvement in diagnostic accuracy of AC images compared with NC images was a result of improved detection specificity (P<.05) with no loss in detection sensitivity. With an angiographic cutoff of ≥50% stenosis, visual interpretation correctly identified 9 of 11 patients without CHD using AC images but only 5 of 11 nondiseased patients when using NC images. For the two false-positive AC studies, one patient had congestive heart failure and a moderately dilated left ventricle, whereas the other had a large severe anterior wall myocardial infarction that extended into the LCx territory. For the latter patient, the LCx was unstenosed, resulting in a false-positive interpretation of that vascular territory. The six false-positive readings from the NC interpretations included the two patients with false-positive readings from the AC interpretations, three patients (all males) with diaphragmatic attenuation (abnormal RCA interpretations), and one female patient with anterior breast attenuation artifact (abnormal LAD interpretation). Similar patterns of abnormality were demonstrated by quantitative analysis.
Although visual interpretation of AC images demonstrated increased sensitivity for the detection of CHD compared with NC images, the increase was not statistically significant at either angiographic stenosis cutoff. Visual interpretation of the AC images with an angiographic cutoff of ≥50% stenosis produced eight false-negative interpretations: three patients had 50% one-vessel stenoses, two patients had two vessels with 50% stenoses, and three patients had balanced three-vessel disease (all vessels ≥70% stenosed). The same patterns were seen in the NC images where 5 of the 11 false-negative interpretations were in patients with mild one-vessel disease (50% to 70% stenosis). The negative interpretations of the images from patients with mild SVD may be due to (1) differences between physiological and angiographic disease severity27 and (2) interobserver variability in the interpretations of coronary angiograms.28 The second most prominent cause of false-negative interpretations was balanced three-vessel disease. Because the interpretation of SPECT perfusion tomograms relies on the relative regional distribution of activity in the myocardium, balanced coronary disease can present as a uniform distribution. For balanced coronary disease, absolute quantification of tracer uptake in the myocardium is necessary to distinguish normal from impaired coronary reserve. With the ability to correct SPECT data for photon attenuation and the future implementation of scatter correction methodology,29 quantification of perfusion may soon be achievable with the potential for increased sensitivity, especially in patients with balanced MVD and patients with relatively mild disease in general.
With an angiographic cutoff of ≥50%, the sensitivity (0.84) and specificity (0.46) from the quantitatively interpreted NC images agreed well with the sensitivity (0.87) and specificity (0.36) values from the multicenter trial for NC stress 99mTc-sestamibi studies reported by Van Train et al.3 From the data presented here, it is apparent that the principle benefit of AC is a significant increase in test specificity (0.82 versus 0.46) with no loss or a slight gain in sensitivity (0.92 versus 0.84).
Localization of CHD
At both angiographic stenoses criteria, diagnostic accuracy in localizing stenosed coronary vessels increased with the interpretation of AC compared with NC images. From the quantitative analyses, the increases in accuracy for disease localization using AC images resulted from large increases in sensitivity. The increased sensitivity can be attributed to the uniformity of the mean and variance distributions of the AC polarmaps compared with the normal NC maps. The uniform distribution of the AC polarmaps provides stricter defect thresholds for the detection of disease, resulting in improved sensitivity with no loss in specificity. From the visual interpretations, increases in accuracy for the localization of stenoses resulted from improvements in vessel sensitivity (LAD and LCx territories) and in specificity (RCA territory). These changes in sensitivity and specificity depend on the observer’s threshold for abnormal; therefore, other observers may show increases in sensitivity and specificity directionally opposite to the results of this study. Regardless of the directional change in sensitivity and specificity, the ROC data predict that diagnostic accuracy will increase for the localization of stenosed vessels using AC images rather than NC images.
Comparison of Visual and Quantitative Results
For the detection of CHD, the visual and quantitative results agreed well. Both demonstrated that AC images more clearly differentiate stenosed from nonstenosed states compared with NC images and that the improvement in diagnostic accuracy results primarily from increased specificity. However, in the localization of stenosed vessels, visual and quantitative interpretations did not agree as well. The discrepancies between visual and quantitative results are attributed primarily to differences in the thresholds used for the detection of disease. For visual analyses, the observer uses a “learned” normal database to set mental limits for the presence or absence of disease. Quantitative analyses used predefined abnormal thresholds for each pixel based on a database of normal studies. Based on these differences, it was expected that sensitivity and specificity pairs might differ between visual and quantitative interpretations. The parameter “accuracy” is less dependent on the abnormality threshold used because it combines the results from the diseased and undiseased states. In comparing the data from Tables 2⇑ and 3⇑, there is good agreement in the accuracy values between visual and quantitative analyses. The same agreement is seen in the ROC curves, which are much more informative regarding the differences between AC and NC images as they provide all sensitivity and specificity pairs for all detection thresholds. The minor differences in the significance of the results between visual and quantitative ROC analyses can be attributed to the statistical uncertainty from a relatively small patient population.
Data Truncation and Iterative Reconstruction
A triple-detector system with a line source positioned opposite a single fan-beam collimator provides an excellent imaging geometry for simultaneous TCT/ECT. With this geometry, cross-talk contamination between the transmission and emission energy windows is directly measured and corrected, minimizing the introduction of bias in the reconstructed images. Patient data are acquired in less than 20 minutes, minimizing artifacts due to patient movement that are more likely to occur with longer acquisitions. With a moveable line source, patient contoured orbits are achievable, reducing the amount of transmission data truncation and maximizing emission image resolution. The single limitation of this system is the truncation of the transmission data due to the fan-beam collimator; this necessitates an iterative reconstruction algorithm to accurately reconstruct the undersampled regions of the thorax. In an effort to decrease the computation time required to provide accurate estimates of the truncated regions of the thorax, we incorporated a quadratic penalty function in the objective function used in the reconstruction of the truncated transmission data.20 The quadratic penalty function constrains the amplitude of reconstructed image pixels based on their nearest neighbors. As a result, the penalty function quickly fills in the missing information based on neighboring pixels. A consequence of the quadratic penalty function is a smoothing effect on high-contrast edges in the reconstructed attenuation maps. Although lung-tissue boundaries have high-contrast edges, respiratory motion during imaging effectively decreases the contrast at these boundaries. The small loss in contrast due to the quadratic penalty function used in the reconstruction of the transmission maps is greatly outweighed by its ability to estimate undersampled regions in the thorax. The incorporation of nonquadratic penalty functions into the reconstruction objective function is the basis of future work.
Limitations of AC Cardiac SPECT
The data from this study have demonstrated the diagnostic improvements that attenuation corrected SPECT with TCT/ECT imaging can offer to myocardial perfusion SPECT. However, from the comparison of AC and NC images, two potential areas of concern have become apparent. The first concern is the 10% to 15% decrease in activity in the apex relative to the uniform activity distribution in the distal and proximal regions of the myocardium of the AC images. This behavior requires the physician interpreting the AC perfusion tomographs to recognize mild reductions in apical activity as normal. Although this internal adjustment by the interpreting physician is similar to the adjustment in the rules regarding diaphragmatic and breast attenuation patterns in NC images, the decreased activity at the apex is a consistent finding not dependent on gender or body habitus. This independence produces less variability in the activity distribution of the AC apex compared with the activity variations resulting from soft tissue attenuation in NC images. As a result, the AC images demonstrated significant improvements in diagnostic accuracy compared with NC images. However, it is important that physicians interpreting AC images familiarize themselves with the AC normal perfusion distribution as the assumption of a homogeneous activity distribution throughout the entire myocardium could lead to false-positive interpretations in the distal LAD territory.
The second potential problem with interpretation of AC images involves splachnic activity, which can spillover into nearby myocardium. The spillover of activity into the myocardium results from photon scatter in the patient and the dispersion of photons due to the distance between the patient and camera. Photon scatter is the more problematic of the two phenomena. Activity from nearby liver and bowel are the most troublesome source of scattered activity. Although photon scatter from the liver has been known to be a problem in the interpretation of the inferoposteroseptal walls of NC perfusion images in some patients, the effect is more severe in AC images. Because the attenuation correction procedure cannot distinguish scattered from unscattered photons, both components are equally amplified during attenuation correction. As a result, the scattered radiation is more prominently displayed in the AC images and poses a larger problem in the interpretation of AC images compared with the NC images of some patients. From the 119 cases reviewed for this article, liver activity was problematic in 6 NC and 13 AC images. To alleviate this problem, scatter subtraction methods need be implemented and validated. We are working on the implementation of an energy deconvolution method for estimating and removing the scatter component from the measured signal.29 To address the problem of photon dispersion, we are implementing three-dimensional iterative reconstruction methods incorporating the depth-dependent response of the imaging system in the system response matrix of our reconstruction algorithms.
Myocardial SPECT perfusion imaging has been criticized for inhomogeneous radiotracer distributions in healthy individuals. This phenomenon has been attributed to photon attenuation and has been blamed for reduced test specificity. We have demonstrated that AC myocardial SPECT perfusion imaging using measured, patient-specific attenuation maps effectively eliminates attenuation artifacts from these studies and significantly increases their specificity for the detection of CHD without a loss in test sensitivity. The localization of disease was also improved, evident by a significant increase in stenosed vessel sensitivity. Based on the increased diagnostic accuracy seen in this study for AC myocardial perfusion SPECT imaging, serious consideration for the clinical implementation of this promising new technology in the diagnosis of CHD is warranted. Intensive further investigation is necessary before routine clinical use.
Selected Abbreviations and Acronyms
|CHD||=||coronary heart disease|
|LAD||=||left anterior descending coronary artery|
|LCx||=||left circumflex artery|
|RCA||=||right coronary artery|
|ROC||=||receiver operating characteristic|
|SPECT||=||single photon emission computed tomography|
|TCT||=||transmission computed tomography|
This work was supported by NIH grant RO1-HL-41047 (Dr Corbett) and NIH grant R29-CA-60711 (Dr Fessler). The authors thank Robert J. Ackermann and Steven Pitt for their help in data acquisition and processing. The authors also wish to thank the Ohio Imaging Division of Picker International for their engineering and software support.
Reprint requests to Edward P. Ficaro, PhD, University of Michigan Hospitals, B1G412/0028, 1500 E Medical Center Dr, Ann Arbor, MI 48109. E-mail firstname.lastname@example.org.
- Received May 30, 1995.
- Revision received September 18, 1995.
- Accepted September 24, 1995.
- Copyright © 1996 by American Heart Association
DePasquale EE, Nody AC, DePuey EG, Garcia EV, Pilcher G, Bredlau C, Roubin G, Gober A, Gruentzig A, D’Amato P, Berger H. Quantitative rotational thallium-201 tomography for identifying and localizing coronary artery disease. Circulation. 1988;77:316-327.
Allman KC, Berry J, Sucharski LA, Stafford KA, Petry NA, Wysor W, Schwaiger M. Determination of extent and location of coronary artery disease in patients without prior myocardial infarction by thallium-201 tomography with pharmacologic stress. J Nucl Med. 1992;33:2067-2073.
Van Train KF, Garcia EV, Maddahi J, Areeda J, Cooke CD, Kiat H, Silagan G, Folks R, Friedman J, Matzer L, Germano G, Bateman T, Ziffer J, DePuey EG, Fink-Bennett D, Cloninger K, Berman DS. Multicenter trial validation for quantitative analysis of same-day rest-stress technetium-99m-sestamibi myocardial tomograms. J Nucl Med. 1994;35:609-618.
Eisner RL, Tamas MJ, Clononger K, Shonkoff D, Oates JA, Gober AM, Dunn DW, Malko JA, Churchwell AL, Patterson RE. Normal SPECT thallium-201 bull’s-eye display: gender differences. J Nucl Med. 1988;29:1901-1909.
DePuey EG, Garcia EV. Optimal specificity of thallium-201 SPECT through recognition of imaging artifacts. J Nucl Med. 1989;30:441-449.
Frey EC, Tsui BM, Perry JR. Simultaneous acquisition of emission and transmission data for improved Tl-201 cardiac SPECT imaging using a Tc-99m transmission source. J Nucl Med. 1992;33:2238-2245.
Jaszczak RJ, Gilland DR, Hanson MW, Jang S, Greer KL, Coleman RE. Fast transmission CT for determining attenuation maps using collimated line source, rotatable air-copper-lead attenuators and fan-beam collimation. J Nucl Med. 1993;34:1577-1586.
Tan P, Bailey DL, Meikle SR, Eberl S, Fulton RR, Hutton BF. A scanning line source for simultaneous emission and transmission measurements in SPECT. J Nucl Med. 1993;34:1752-1760.
Ficaro EP, Fessler JA, Rogers WL, Schwaiger M. Comparison of Am-241 and Tc-99m as transmission source for the attenuation correction of Tl-201 SPECT imaging of the heart. J Nucl Med. 1994;35:652-663.
Ficaro EP, Fessler JA, Ackermann RJ, Rogers WL, Corbett JR, Schwaiger M. Simultaneous transmission-emission thallium-201 cardiac SPECT: effect of attenuation correction on myocardial tracer distribution. J Nucl Med. 1995;36:921-931.
Verani MS. Dobutamine myocardial perfusion imaging. J Nucl Med. 1994;35:737-739.
Sauer K, Bouman C. A local update strategy for iterative reconstruction from projections. IEEE Trans Signal Proc. 1993;41:523-548.
Fessler JA. Penalized weighted least-squares image reconstruction for positron emission tomography. IEEE Trans Med Imaging. 1994;13:290-300.
Garcia EV, Van Train K, Maddahi J, Prigent F, Friedman J, Areeda J, Waxman A, Berman DS. Quantification of rotational thallium-201 myocardial tomography. J Nucl Med. 1985;26:17-26.
Maddahi J, Van Train K, Prigent F, Garcia EV, Friedman J, Ostrzega E, Berman D. Quantitative single photon emission computed thallium-201 tomography for detection and localization of coronary artery disease: optimization and prospective validation of a new technique. J Am Coll Cardiol. 1989;14:1689-1699.
Metz CE, Wang P, Kronman HB. A new approach for testing the significance of differences between ROC curves measured from correlated data. In: Deconinck F, ed. Information Processing in Medical Imaging. The Hague, Netherlands: Martinus Nijhoff; 1984:432-445.
Zir LM, Miller SW, Dinsmore RE, Gilbert JP, Harthorne JW. Interobserver variability in coronary angiography. Circulation. 1976;53:627-632.