Optimization and evaluation of 4D MAP-RBI-EM image reconstruction method for gated myocardial perfusion SPECT

The objective of this research was to investigate the parameters of generalized 4D space-time Gibbs priors used in 4D MAP-RBI-EM image reconstruction method and to develop a task-based human observer study to evaluate and optimize the method for application to gated myocardial perfusion (GMP) SPECT. To study the characteristics of the parameters, we created simulated GMP SPECT data from a realistic 4D NCAT phantom modeling the uptake distribution of Tc-99m Sestamibi. Noise-free and noisy projection data were generated using an analytical projector that included the effects of attenuation, collimator-detector response and scatter (ADS). The normalized mean squared error (NMSE) in noise-free images and the averaged normalized standard deviation (NSDav) in noisy images were used to describe the spatial and temporal image fidelity with respect to its true phantom image, and image noise magnitude, respectively. We postulated that the optimal value ranges for each parameter of the image reconstruction methods for best image quality were within the region where the trade off was minimum for both NMSE and the NSDav. The results showed that the 4D MAP-RBI-EM with parameters in their own optimal range and correction of image degrading factors provides improved reconstructed image quality as compared to 3D FBP method followed by 4D post-filtering or 3D OS-EM method with the same correction followed by 4D post-filtering. Based on these results, we used a task-based human observer study to evaluate the proposed method and to find the optimal NMSE-NSDav trade-off for the detection task of abnormal myocardial wall motion. Using the 4D NCAT phantom, we created a realistic hypokinetic cardiac motion model as well as a normal motion model. The projection data were generated in the same way as above, and reconstructed using the 3D FBP and the 3D OS-EM methods with ADS corrections followed by different cut-off frequencies of the 4D post-filter to include the reconstructed images from a wide range of the NMSE-NSDav trade-off for the observer study. 4D MAP-RBI-EM method with ADS corrections was also used to reconstruct the projection data using different combinations of prior parameters. Using a newly developed motion image display application for human observer studies, observers were trained to the multi-sliced and multi-oriented gated SPECT images animated with a realistic real-time frame rate, and were instructed to rate their confidence on the absence or presence of a regional myocardial motion defect on a continuous scale. We applied receiver operating characteristic (ROC) analysis and used the area under the ROC curve (AUC) as an index of performance for comparison. The result showed that 4D MAP-RBI-EM with ADS correction had the highest AUC value, resulting in significantly better human observer detection performance among the tested methods, and that the optimal NMSE-NSDav trade-off from optimized reconstruction parameters corresponded to the maximum AUC value for each method. We found that the NMSE-NSD av trade-off could be used as a useful guide in choosing the optimal range of parameter values for each method, and conclude that the optimized 4D MAP-RBI-EM method with corrections for image degrading factors provides significant improvement in detecting wall motion abnormalities with GMP SPECT.