Application of a generic motion model for PET respiratory motion correction

International audienceObjectives: Respiratory motion correction using elastic transformations applied to list-mode data during image reconstruction can lead to motion corrected PET images. The objective of this work is the use of a generic respiratory motion model to derive the necessary deformations for incorporation during the PET reconstruction process. Methods: The generic model used is based on principal component analysis (PCA). The fitting of the model to a given patients anatomy requires only two static CT images (for example end-inspiration and end-expiration) in combination with respiratory synchronized images of the patients body surface. Our generic model computes a relation between the surface map acquired by an external camera and the internal motion described by the deformation fields. The resulting patient adapted motion model is subsequently used to generate n CT images (n=6, 11, 21 and 41 frames, with n=6 being equivalent to commonly acquired 4D CT series) and corresponding n-1 deformation matrices for the correction process. In this study 4 patients were used to create the generic model which was subsequently tested on two different patients. The intrinsic global model efficiency was firstly tested by generating 4D CT series and comparing it to the corresponding acquired 4D CT. Results: Good correlation (r=0,93 ± 0,4) and expert validation (1.9 mm ± 0.6 mm difference based on the identification of anatomical landmarks) was found between the acquired 4D CT series and the model generated images in the two patients tested. Qualitative and quantitative measures demonstrated that using 21 generated images (20 deformation matrices) leads to superior motion correction compared to the use of 5 or 10 deformation matrices. Using a finer temporal sampling did not improve the results any further. Conclusions: The use of a generic motion model for PET image respiratory motion correction avoids unnecessary 4D CT acquisitions and improves the temporal resolution of the deformation matrices used in the correction process with an associated improvement in the PET corrected images