Respiratory Motion Estimation Using Vibration Model and Reducing Motion Blur through Deconvolution: A Simulation Study

摘要

Respiratory motion is a main source of degradation in positron emission tomography (PET)/computed tomography (CT) image, it leads to significant motion artifacts of PET image which are known to influence diagnosis and treatments in radiation oncology. Existing approaches to correct motion artifacts involve using gating devices and/or 4D CT. However they either suffer from high CT dose or high computation burden. In this paper we present a sinusoid vibration model to simulate the respiratory motion, the motion extent and direction are derived from Radon transform of the cepstrum of blurred image. Then we employ three typical deconvolution algorithms (Wiener filter, Constrained least square, and Richardson-Lucy) to eliminate the motion blur respectively according to estimated parameters and compare their de-blurring results. The experiments on both synthetic and phantom images show good performance of our method in identifying vibration modeled respiration motion. The advantage of our method lies in its convenience and economy since only static PET image is needed for analysis.