PET/CT Motion Correction Exploiting Motion Models Fit on Coarsely Gated Data Applied to Finely Gated Data

Motion correction is imperative to the reduction of blurring and artefacts inherent in PET; due to the relatively long acquisition time, and the temporal difference in the attenuation map acquisition. Registration literature contains many examples of spatial regularisers, however there are few temporal regularisers. Motion models can act as such a temporal regulariser, as well as allowing for the interpolation of unseen motion correction results. In our previous work, we applied a motion modelling approach to high TOF resolution non-attenuation corrected data; where the data was corrected to the space of the attenuation map. However, this approach was challenging, especially when low contrast lung tumours are present. This work seeks to extend previous work, by incorporating an approach suggested by Y. Lu et al. (JNM 2018), to perform an initial MLACF reconstruction for the motion estimation. In this work, we combine these two approaches, with several improvements, including; µ-map alignment, as well as, fitting the motion model on low noise low temporal/gate resolution data, and applying it to high noise high temporal/gate resolution data. To test this, XCAT volumes are constructed, and TOF data simulated. Evaluation compares the results of the proposed method against, where the motion model was fit on data gated more finely, where the motion model was fit on noiseless data, and finally non-motion corrected examples. Results indicate that the incorporation of MLACF, and fitting of the motion model on low noise low temporal/gate resolution data, improves contrast and quantification, while allowing for a relatively fast execution time.