Motion Correction for Dynamic PET/MR Datasets

Abstract

Abstract Positron emission tomography (PET) is an imaging technique that uses short lived radioactive labelled molecules to visualise and measure physiological activities. Dynamic PET acquisition requires a long scan time, and movement of the subject during this time cannot be fully avoided. Motion of organs due to breathing, heartbeat, etc. is continuous, but there are also bulk motions of body regions to consider which may occur if the subject has difficulty laying still. Motion of subjects during a PET scan is an important factor that degrades image resolution and quality, and thereby limits its potential to provide accurate readouts. Therefore, this study aimed to implement an approach for motion correction of a dynamic whole-body hybrid PET/MR dataset. The method consisted of utilising a deformable image registration in a post-reconstruction approach on the MR images to find optimal displacement fields, and then applying these on the PET images. The registration method used in this study was fast graph-cut based optimisation, which was evaluated with Dice score and properties of the displacement field. The transformed PET images were evaluated with Patlak modelling and visualised with segmentations. Results regarding the image registration part show an improvement in terms of Dice score and reasonable properties of the displacement fields. For the motion corrected PET images there is a visual difference for images containing considerable motion. There is also a decrease in error of the Patlak slope fit in the Patlak modelling, indicating an improvement of the correct localisation of the segments over time. The conclusion is that the used method gives promising results for the studied dataset, however further research is needed with larger datasets that also contain a wider range of movements.