Evaluation and Optimization of Motion Correction in Spinal Cord fMRI Preprocessing

Abstract

Motion correction is an essential step in the preprocessing of functional magnetic resonance imaging (fMRI) data, improving the temporal signal to noise ratio (tSNR) and removing unwanted variance. Because of the characteristics of the spinal cord (non-rigidity, surrounded by moving organs), motion correction becomes especially challenging. We compared the efficiency of different motion correction protocols and suggest a preferred method for spinal cord fMRI data. Here we acquired gradient-echo echo-planar-imaging axial lumbar spinal cord fMRI data during painful mechanical stimulation of the left lower extremity of 15 healthy volunteers on a 3T scanner. We compared multiple motion correction techniques: 2D and 3D FLIRT realignment with and without slice-wise regulation, SliceCorr (implemented in the Spinal Cord Toolbox) and proposed a method 3D FLIRT in addition to Slice Regulation (SLiceReg) along the spinal cord. TSNR, image entropy, DVARS, image Sum of Absolute Differences and number of activated voxels in the spinal cord from GLM analysis to evaluate the performance of multiple motion correction procedures. The tSNR and DVARS 3D FLIRT + SLiceReg were significantly improved over other realignment methods (p<0.001). In comparison, tSNR=14.20{+/-}0.02 and DVARS=165.77{+/-}1.54 were higher than other methods. Additionally, the number of activated voxels of the statistical map in our suggested method was higher than the other realignment methods (p<0.05). Our results illustrated the proposed motion correction algorithm that integrated 3D motion correction and 2D slicewise regularization along spinal cord curvature could improve subject-level processing outputs by reducing the effects of motions. Our proposed protocols can improve subject-level analysis, especially in lumbar region that suffers from involuntary motions and signal loss due to susceptibility effect more than other spinal cord regions.