Evaluation of 3D Monte Carlo-based scatter correction for 201Tl cardiac perfusion SPECT.

UNLABELLED (201)Tl-Chloride ((201)Tl) is a myocardial perfusion SPECT agent with excellent biochemical properties commonly used for assessing tissue viability. However, cardiac (201)Tl SPECT images are severely degraded by photons scattered in the thorax. Accurate correction for this scatter is complicated by the nonuniform density and varied sizes of thoraxes, by the additional attenuation and scatter caused by female patients' breasts, and by the energy spectrum of (201)Tl. Monte Carlo simulation is a general and accurate method well suited to modeling this scatter. METHODS Statistical reconstruction that includes Monte Carlo modeling of scatter was compared with statistical reconstruction algorithms not corrected for scatter. In the ADS method, corrections for attenuation, detector response, and scatter (Monte Carlo-based) were implemented simultaneously via the dual-matrix ordered-subset expectation maximization algorithm with a Monte Carlo simulator as part of the forward projector. The ADS method was compared with the A method (ordered-subset expectation maximization with attenuation correction) and with the AD method (a method like the A method but with detector response modeling added). A dual-head SPECT system equipped with two (153)Gd scanning line sources was used for simultaneously acquiring transmission and emission data. Four clinically realistic phantom configurations (a large thorax and a small thorax, each with and without breasts) with a cardiac insert containing 2 cold defects were used to evaluate the proposed reconstruction algorithms. We compared the performance of the different algorithms in terms of noise properties, contrast-to-noise ratios, the contrast separability of perfusion defects, uniformity, and robustness to anatomic variations. RESULTS The ADS method provided images with clearly better visual defect contrast than did the other methods. The contrasts achieved with the ADS method were 10%-24% higher than those achieved with the AD method and 11%-37% higher than those achieved with the A method. For a typical contrast level, the ADS method exhibited noise levels around 27% lower than the AD method and 34% lower than the A method. Compared with the other 2 algorithms, the ADS reconstructions were less sensitive to anatomic variations and had better image uniformity in the homogeneously perfused myocardium. Finally, we found that the improvements that can be achieved with Monte Carlo-based scatter correction are stronger for (201)Tl than for (99m)Tc imaging. CONCLUSION Our results indicate that Monte Carlo-based scatter correction is suitable for (201)Tl cardiac imaging and that such correction simultaneously improves several image-quality metrics.