Maximum Likelihood Preprocessing for Improved Filtered Back‐Projection Reconstructions

Objective Based upon theoretical consideration and experimental investigation, we propose a model-based maximum likelihood (ML) preprocessing approach to the filtered back-projection (FBP) algorithm for application to single photon emission CT (SPECT). Materials and Methods In this approach, we optimally removed Poisson noise and an average blur from the projection images and then applied the FBP algorithm for reconstruction. Results The average blur was due to the combined effect of the geometric response of the imaging system, septal penetration, scatter, and patient motion. We evaluated this preprocessing approach to FBP reconstruction on SPECT physical phantom and patient studies. Preliminary results were encouraging. Conclusion The advantages of the proposed approach are that statistical noise is accurately modeled, removal of both blur and noise is optimal in the ML sense, and further improvement in this method may be achieved through more accurate modeling of blur in the projection images.