Automated multi-resolution brain motion correction for enabling accurate quantification in dynamic PET non-lesional epilepsy studies

Abstract-Text (inklusive Referenzen und Bildunterschriften)

Aim

Patient head motion has both quantitative and qualitative impacts on PET images, in particular, for dynamic brain imaging. We propose a fast, fully automated brain motion correction technique based on multiscale registration to perform accurate motion compensation for absolute quantification studies in non-lesional epilepsy patients.

Methods

15 patients with drug-resistant non-lesional extratemporal lobe epilepsies underwent an 18F-FDG PET brain examination in a fully integrated PET/MRI. The imaging protocol consisted of a 60 min PET list-mode acquisition, with parallel MR-navigator acquisitions for tracking motion. PET list-mode data was re-binned into a dynamic frame sequence (24 x 5 s, 1 x 60 s, 1 x 120 s, 11 x 300 s) and was reconstructed using the Ordinary Poisson Ordered Subset Expectation-Maximization OP-OSEM 3D algorithm. Attenuation and scatter correction were performed using AC-maps corrected for motion. The proposed motion correction performs automated co-registration in multiple scales, starting alignment between images at the lowest scale and is repeated until it reaches the finest possible scale. To quantify the accuracy of the proposed method, we used absolute difference as the similarity metric, comparing the motion vectors obtained from the multiscale approach directly to the motion vectors obtained from the MR navigators.

Results

The absolute difference between the motion vectors obtained from the MR navigators and the multiscale registration approach was less than five percent for frames later than 2 min of post-injection (p.i). However, the multiscale approach was not able to calculate the motion vectors for frames less than 2 min p.i, as there was no sufficient structural information to start with.

Conclusion

We propose a fully-automated motion compensation approach for 18F-FDG PET brain studies based on a multi-resolution scheme. The proposed motion compensation technique successfully accounts for patient motion, therefore enabling accurate non-invasive kinetic modelling studies in non-lesional epilepsy patients.