An investigation of spatial resolution and noise in reconstructed CT images using iterative reconstruction (IR) and filtered back-projection (FBP)

This study compared the spatial resolution and the amount of noise in CT images reconstructed using the iterative reconstruction (IR) and filtered back-projection (FBP) techniques. We used a circular homogeneous phantom (pixel value of 0.5) containing a small circle at its center, with a diameter of 0.53-cm and a pixel value of 1.0. In order to reconstruct the image, we used the simultaneous iterative reconstruction technique (SIRT) with 5, 15, 25, 35 and 45 iterations respectively, and FBP with a Ram-Lak filter. The spatial resolution of the reconstructed image was characterized by the modulation transfer function (MTF) and image noise was characterized by the standard deviation within the homogeneous area. We found that the spatial resolution using SIRT with 35 iterations is similar to using FBP. The spatial resolution using SIRT tends to be worse using less than 35 iterations, and it tends to be better using more than 35 iterations. Using SIRT, image noise decreased with increasing iterations. The standard deviation for the SIRT technique using 45 iterations was about 0.02 similar to the FBP technique. Thus, SIRT could potentially produce better spatial resolution than the FBP technique with comparable image noise. (paper)