Bayesian SPECT lung imaging for visualization and quantification of pulmonary perfusion

The authors quantitatively and qualitatively examine the use of a Gibbs prior in maximum a posteriori (MAP) reconstruction of SPECT images of pulmonary perfusion using the expectation-maximization algorithm (EM). This Bayesian approach is applied to SPECT projection data acquired from a realistic torso phantom with spherical defects in the lungs simulating perfusion deficits. Both the scatter subtraction constant (k) and the smoothing parameter beta (β) characterizing the prior are varied to study their affect on image quality and quantification. Region of interest (ROI) analysis is used to compare MAP-EM radionuclide concentration estimates with those derived from a “clinical” implementation of filtered backprojection (CFBP), and a quantitative implementation of FBP (QFBP) utilizing nonuniform attenuation and scatter compensation. Qualitatively, the MAP-EM images contain reduced artifacts near the lung boundaries relative to the FBP implementations. Overall MAP-FM image visual quality and the ability to discern the areas of reduced radionuclide concentration in the lungs depend on the value of β and the total number of iterations. For certain choices of β and total iterations, MAP-EM lung images are visually comparable to FBP. Based on profile and ROI analysis, SPECT QFBP and MAP-FM images have the potential to provide quantitatively accurate reconstructions when compared to CFBP; computational burden, however, is greater for the MAP-FM approach. To demonstrate the potential clinical efficacy of the methods, the authors present pulmonary perfusion images of a patient with lung cancer