Recursive Bayesian estimation for respiratory motion correction in Nuclear Medicine imaging

Respiratory motion correction degrades quantitatively and qualitatively Nuclear Medicine images. We propose that adaptive approaches are required to correct for the irregular breathing patterns often encountered in the clinical setting, which can be addressed within a Bayesian tracking formulation. This allows inference of the hidden organ configurations using only knowledge of an external observation such as a parametrized external surface. The flexible framework described provides a method to correct for organ motion whilst accommodating for irregular unseen respiratory patterns. In this work we utilize a Kalman filter and compare it with a Particle filter. A novel adaptive state transition model is also introduced to describe the evolution of organ configurations. The Kalman filter marginally outperforms the Particle filter, both approaches however offer an effective motion correction mechanism, correcting for motion with errors of around 1-3mm. We present results of simulated PET images derived from XCAT to demonstrate the efficacy of the approach.