Rapidly converging iterative reconstruction algorithms in single-photon emission computed tomography

Iterative reconstruction algorithms with markedly different convergence rates have been proposed in single-photon emission computed tomography (SPECT). Several new iterative reconstruction methods are described in this investigation. Differences between the methods include whether a ramp filter was used during backprojection, the type of backprojection weighting and whether camera and collimator blur were employed in the projection step. Simulated and real cylindrical phantoms with rod inserts were used to compare the properties of convergence and resolution following reconstruction by maximum likelihood (ML), iterative-Chang and the newly proposed reconstruction methods. Resolution was assessed after kernel-sieve regularization to achieve the same signal-to-noise ratio for all methods. Compared with maximum-likelihood reconstruction, methods employing a ramp converged much faster. One such method resulted in images with the same resolution and noise as ML, thus permitting termination of reconstruction at 14 iterations rather than the 1000 iterations required with ML. The major determinants of resolution were found to be use of an accurate model of the gamma camera imaging process in the projection step and inclusion of attenuation weighting and depth-dependent blur in the backprojection step. In summary, a new iterative reconstruction method was developed incorporating attenuation and blur and using a ramp filter that achieved results comparable to maximum-likelihood reconstruction in a fraction of the time.