Brunner, M. E. MS; Eisner, R. L. PhD; Coumans, J. J. PhD; Fajman, W. A. MD; Flower, S. MD; Garrett, J. C. MD Author Information

A method of reconstruction of single-photon emission computed tomographic (SPECT) images--distance-weighted filtered backprojection--is described. The quality of SPECT images using the standard filtered backprojection algorithm for data obtained from 360 degrees acquisitions is limited by the loss of spatial resolution with distance, and the attenuation and scatter inherent in the planar images used for reconstruction. In contrast to the standard techniques in which a reconstructed pixel receives equal contributions from 180 degrees opposed views, distance--weighted filtered backprojection applies a variable weighting factor to the data such that the pixel receives greater weight from the closest planar views. Data derived from hot and cold spot phantoms as well as from various clinical studies show better spatial and contrast resolution with this new technique compared with the conventional 360 degrees algorithm.

A simple analytic model is presented which is used to calculate the impulse response of a computed tomography system for parametric variations of the geometry. The model is useful for fan beam as well as parallel beam geometries. Results show that wide-aperture detector systems have intrinsic resolution limitations that are not mitigated by oversampling. Furthermore, the optimum focal spot size of the X-ray source is found to be dependent on the detector geometry.

In computed tomography, each projection, or view, is broken up into n data samples. If the interval between samples is d there is an associated Nyquist frequency KN = 1/(2d), which represents the highest spatial frequency retrievable after the sampling process. If higher frequencies are present, as in the case of sharp edges of bone, they will cause aliasing errors which show up on the final image as streaks radiating from the edges. These streaks differ from interpolation streaks in that they are not synchronous with the view angles and are not affected by the number of views. Three remedies for aliasing streaks are (i) broadening the source-detector aperture function so as to physically filter out the high frequencies before sampling: (ii) finer sampling of the data, followed by digital filtering to remove the high frequencies (attractive with translate-rotate or stationary detector scanners); or (iii) offsetting the detector sampling points (attractive with rotating-detector 360° scanners). With remedy iii. opposing views can either be interleaved, to double the spatial resolution, or processed separately, in which case the aliasing error cancels. These remedies are illustrated for both parallel and fan-beam projections. In the case of fan beam projections, the cancellation is close but not exact, because opposing rays do not precisely coincide.