Estimation of Physiological Motion Using Highly Accelerated Continuous 2D MRI

Abstract

Patient motion is well-known for degrading image quality during medical imaging. Especially positron emission tomography (PET) is susceptible to motion due to its usually long scan times. In hybrid PET/MRI (magnetic resonance imaging), simultaneously acquired dynamic MRI data can be used to correct for motion. Usually, MRI model-based motion correction approaches are applied to the PET data. However, these approaches may fail for nonpredictable, irregular motion. We propose a novel approach for the continuous and real-time tracking of motion using highly accelerated, dynamic MRI for an accurate motion estimation. For this purpose, a TurboFLASH sequence is utilized in single-shot mode with additional exploiting GRAPPA acceleration. Sampling frequency for one slice is up to 26 ms and 520 ms for one 3D volume of 20 coronal slices. Principal component analysis and a phase-sensitive resorting of slices is performed to restore temporal consistency of the volumes. Motion is estimated from these volumes using hyper-elastic registration. The approach is validated with the help of a dynamic thorax phantom as well as with eleven healthy volunteers. Phantom ground-truth data demonstrates that the approach produces an accurate motion estimation. Volunteer validation proves that the approach is also valid for different respiratory amplitudes including highly irregular breathing. The approach could be