Image-based modelling of residual blurring in motion corrected small animal PET imaging using motion dependent point spread functions

Abstract

Optical motion tracking and motion compensation reconstruction algorithms enable the acquisition of quantitative measurements of brain function on conscious and freely moving rodents. However, motion corrected images often exhibit reduced resolution when compared with their stationary counterparts. This apparent loss of resolution can be attributed, among others, to jitter/noise in the measured motion estimates and brief periods of fast animal motion with insufficient motion sampling rate. In this paper we propose a novel methodology to experimentally characterise the residual blurring in the motion corrected images by measuring the motion-dependent point spread function (PSF) in image space using a point source rigidly attached on the moving object. We evaluated the proposed methodology using experimental phantom measurements acquired on the microPET Focus220 scanner. The motion dependent point spread function was extracted from the point source attached on the moving phantom, after motion correcting the images and modelling the point source in image space using an Expectation Maximisation algorithm as a weighted sum of two Gaussian distributions. Finally, the fitted blurring kernels were used within an iterative Lucy-Richardson algorithm to mitigate the deblurring in the motion corrected images. For motion typically encountered in an awake rat study, results showed that unprocessed motion corrected images suffer from lower resolution compared to a stationary acquisition. The shape of the measured blurring kernel, correlated well with the motion trajectory, while the width of the kernel was proportional to the speed/acceleration of the object. Post-processed images using the corresponding motion dependent blurring kernel appeared not only qualitatively, but also quantitatively (in terms of contrast) more similar to their stationary counterpart. We conclude that it is possible to experimentally measure the residual motion-dependent blurring kernel and use it within a post reconstruction deconvolution framework to improve resolution and quantification of motion corrected images.