Mitigation of Motion Artifacts in Functional MRI: A Combined Acquisition, Reconstruction and Post Processing Approach.

Abstract

Head motion limits the accuracy, specificity and sensitivity of fMRI. Rigid body registration of fMRI data only corrects for bulk movements while leaving secondary motion artifacts from spin history effects, dynamic field inhomogeneity changes and interpolation errors untouched. Secondary artifacts reduce accuracy of image registration, increase variance in fMRI time-series and reduce sensitivity of detection of active voxels. In this thesis, some approaches to increase robustness of fMRI to head motion have been presented. These involve explicit optimization of acquisition parameters, use of image acquisition and reconstruction methods that reduce secondary motion artifacts and, better isolation and removal of residual motion artifacts that remain after image realignment. Specifically, methods to mitigate motion artifacts include use of thinner slices and slices of variable thickness during image acquisition for better signal recovery in brain regions with large intra-voxel dephasing induced signal loss. A combined forward and reverse spiral k-space trajectory was used to reduce susceptibility artifacts in presence of motion. Iterative image reconstruction with dynamically updated fieldmaps was used to correct temporally changing field inhomogeneity from motion and susceptibility interactions. Results demonstrated that these corrective measures increased the overall robustness of fMRI to susceptibility induced field inhomogeneity, head motion, and dynamic interactions between them. Consequently, better quality of fMRI data also improved the quality of motion correction, reduced variance in the time-series and increased sensitivity of detection of active voxels during fMRI experiments with head movement. Constrained Independent Component Analysis (cICA) was used for modeling, isolation and removal of residual motion artifacts that remain in fMRI time-series despite image registration. cICA was found to be better able to isolate the residual errors compared to the prevalent General Linear Model (GLM) methods. Further, cICA automated the identification and removal of erroneous components and eliminated human errors during this process. Using a combined approach, i.e., by optimizing acquisition parameters, acquisition methods, and reconstruction methods during data collection to improve image quality and motion correction and, by better modeling, isolation and removal of residual motion artifacts using cICA, the impact of head motion on fMRI studies can be vastly reduced.