Pixel driven implementation of filtered backprojection for reconstruction of fan beam SPECT data using a position dependent effective projection bin length

Abstract

Filtered backprojection is commonly implemented as a pixel driven algorithm in which the density is reconstructed on an array of grid points that are usually associated with the centres of square image pixels. In fan beam geometry, this conventional pixel driven approach using two-bin linear interpolation leads to an inefficient use of the projection data due to magnification at the projection line. For typical count limited SPECT data, this results in increased reconstructed image noise. We propose an alternative type of pixel driven algorithm that makes efficient use of projection data by averaging over all bins within a position dependent `effective' projection bin interval. The effective bin interval is equivalent to the image pixel length magnified on the projection line. This heuristic method leads to more complete use of the projection data and results in reconstructed images with superior noise properties compared with the conventional method while presenting similar spatial resolution characteristics.