Regularized B-spline deformable registration for respiratory motion correction in PET images

Abstract

A major challenge in respiratory motion correction of gated PET images is their low signal to noise ratios (SNR). This particularly affects the accuracy of image registration. This paper presents an approach to overcoming this problem using a deformable registration algorithm which is regularized using a Markov random field (MRF). The deformation field is represented using B-splines and is assumed to form a MRF. A regularizer is then derived and introduced to the registration, which penalizes noisy deformation fields. Gated PET images are aligned using this registration algorithm and summed. Experiments with simulated data show that the regularizer effectively suppresses the noise in PET images, yielding satisfactory deformation fields. After motion correction, the PET images have significantly better image quality.