A Bayesian iterative transmission gradient reconstruction algorithm for cardiac SPECT attenuation correction

Abstract

Background: High-quality attenuation maps are critical for attenuation correction of myocardial perfusion single photon emission computed tomography studies. The filtered backprojection (FBP) approach can introduce errors, especially with low-count transmission data. We present a new method for attenuation map reconstruction and examine its performance in phantom and patient data.

Methods and results: The Bayesian iterative transmission gradient algorithm incorporates a spatially varying gamma prior function that preferentially weights estimated attenuation coefficients toward the soft-tissue value while allowing data-driven solutions for lung and bone regions. The performance with attenuation-corrected technetium 99m sestamibi clinical images was evaluated in phantom studies and in 50 low-likelihood patients grouped by body mass index (BMI). The algorithm converged in 15 iterations in the phantom studies. For the clinical studies, soft-tissue estimates had significantly greater uniformity of mediastinal coefficients (mean SD, 0.005 cm(-1) vs 0.011 cm(-1); P < .0001). The accuracy and uniformity of the Bayesian iterative transmission gradient algorithm were independent of BMI, whereas both declined at higher BMI values with FBP. Attenuation-corrected perfusion images showed improvement in myocardial wall variability (4.8% to 4.1%, P = .02) for all BMI groups with the new method compared with FBP.

Conclusion: This new method for attenuation map reconstruction provides rapidly converging and accurate attenuation maps over a wide spectrum of patient BMI values and significantly improves attenuation-corrected perfusion images.