Impact of truncation to the reconstruction with a small number of projections

Dedicated cardiac SPECT systems have the advantage of high speed and sensitivity at no loss or even a gain in resolution. The purpose of this paper is to investigate segmented parallel-hole collimation, develop a tailored Maximum-Likelihood Expectation-Maximization (ML-EM) algorithm for reconstructing truncated projections, and find the minimal number of view-angles required. We simulated and compared three image configurations, including two segmented-parallel-hole detectors based on the conventional and large detectors, and one small multi-detector. To study the effects of truncation and the number of angular sampling, the number of view-angles was varied from 10 to 60 and the truncation degree increased from 91% to 95%. For each view-angle, the root-mean-square error (RMSE) between the reconstructed image and the phantom was calculated and used to give a measure of the accuracy of the reconstructions. Distortion was observed from the shape difference maps of the heart. A tailored MLEM algorithm was used in the reconstruction, wherein the true-projection to forward-projection ratios outside of FOV were set to one instead of zero before backprojection. Finally, an NCAT phantom and GATE Monte Carlo simulation tool were used to verify the conclusions. Computer simulation results show that truncation has little effect on the reconstructed images when sufficient view angles are provided, while it affects the reconstruction with few number of views in terms of reducing contrast and introducing bias to the reconstructions. Based on visual inspection only, 10 view-angles is sufficient to reconstruct the image without shape distortion. To reconstruct both value and shape correctly, 20 view-angles is the minimum requirement for truncated projections. Two segmented conventional detectors are sufficient to reconstruct the image without distortion. Uniform angular sampling gives a considerable improvement in the image quality, but a small rotation is required.