Effects of sinogram filtering in the quality of PET reconstructions: preliminary results

Abstract

Low-pass filtering sinograms prior to reconstruction is a general practice to reduce noise. Sinograms are generally filtered in the radial direction, although other filtering schemes have been proposed. It is known that the Fourier transform of a sinogram shows a particular shape of the spectral energy distribution (“bow-tie”). In this work, this property has been exploited to perform an adapted filter, whose performance has been compared with previously reported methods: angular, axial and stackgram domain filtering. Stackgram and angular filtering degraded resolution (~16 and ~5.7 % respectively) while no significant enhancement in contrast to noise ratio (CNR) was achieved. Angular filtering resulted in a circle blurring artifact dependant on the distance to the center of the FOV. Bow-tie filtering showed the best results (enhancement of ~26% in resolution and of ~12% in CNR). Axial filtering degraded resolution but enhanced CNR (~14 %), appearing as a good strategy to reduce radial filtering. Experiments on rodent images showed a noticeable image quality enhancement achieved when using bow-tie filtering combined with radial and axial filters.