Human-observer receiver-operating-characteristic evaluation of attenuation, scatter, and resolution compensation strategies for (99m)Tc myocardial perfusion imaging

Abstract

Nonuniform attenuation, scatter, and distance-dependent resolution are confounding factors inherent in SPECT imaging. Iterative reconstruction algorithms permit modeling and compensation of these degradations. We investigated through human-observer receiver-operating-characteristic (ROC) studies which (if any) combination of such compensation strategies best improves the accuracy of detection of coronary artery disease (CAD) when expert readers have only stress images for diagnosis. METHODS: A 3-headed SPECT system fitted with a (153)Gd line source was used to acquire simultaneously (99m)Tc-methoxyisobutylisonitrile (MIBI) images and transmission data. With these acquisitions, the accuracy of detecting CAD was evaluated for the following reconstruction strategies: filtered backprojection (FBP); ordered-subset expectation maximization (OSEM) with attenuation correction (AC); OSEM with AC and scatter correction (SC) (AC + SC); and OSEM with AC, SC, and resolution compensation (RC) (AC + SC + RC). Reconstruction parameters for OSEM were optimized by use of human-observer ROC studies with hybrid images, whereas standard clinical parameters were used for FBP. A total of 100 patients, including 55 patients referred for angiography and 45 patients with <5% likelihood for CAD, were included in the ROC studies. Images reconstructed with the 4 methods were rated independently with regard to the presence of CAD by 7 observers using a continuous scale for certainty. RESULTS: With area under the ROC curve (A(z)) as the criterion, the iterative reconstructions with compensation strategies (AC, AC + SC, and AC + SC + RC) demonstrated better detection accuracy than did FBP reconstructions for the overall detection of CAD as well as for the localization of perfusion defects in the 3 vascular territories. In general, the trend was for an increase in the A(z) for the progression from FBP to OSEM with AC, to OSEM with AC + SC, and to OSEM with AC + SC + RC. Statistically, the combination strategy with AC + SC + RC provided significantly higher A(z) values than did FBP images for the overall detection of CAD and the localization of perfusion defects in the left anterior descending coronary artery and left circumflex coronary artery territories, whereas AC + SC provided significantly better performance in the right coronary artery territory. CONCLUSION: The results indicate that OSEM with AC + SC + RC outperforms FBP reconstructions, indicating that the modeling of physical degradations can improve the accuracy of detection of CAD with cardiac perfusion SPECT reconstructions.