Evaluation of motion censoring in fMRI data using inter-/intra-volume motion corrupted MR data
Poster No:
1955
Submission Type:
Abstract Submission
Authors:
Wanyong Shin1, Mark Lowe1
Institutions:
1The Cleveland Clinic, Cleveland, OH
First Author:
Co-Author:
Introduction:
Head motion is the one of the main sources of bias sources rs-/fMRI analysis (1-4). Various motion indexes have been proposed and used to apply thresholds to "censor" motion corrupted volumes in fMRI dataset. In this study, we modify simulated Prospective Acquisition CorrEcted (PACE) EPI sequence (5) and inject different inter- (volume-level) and intra-volume (slice-level) motion to ex-vivo brain phantom during PACE EPI acquisition, referred as Simulated PACE (SIMPACE) data(6). We test which motion index predicts bad motion corrupted volume effectively and how motion censoring improves fMRI dataset using SIMPACE data.
Methods:
We built ex-vivo brain phantom (7) and scanned it using SIMPACE sequence (TR/TE=2s/28ms, voxel size = 2x2x4mm3, 21 slices, 150 vols) with inter-/intra-volume motion patterns injected which were estimated from 10 HCP YA protocol dataset.
Two different motion correction and pipelines were used. SIMPACE data was 3d rigid volume motion corrected (VOLMOCO). After VOLMOCO, 6 rigid motion parameters were regressed out with voxelwise partial volume (PV) nuisance regressors (8). SLOMOCO was applied with 12 volume-/slice-wise motion nuisance and PV regressor (6, 9). We calculate temporal standard deviation (SD) of the residual signal in VOL/SLOMOCO SIMPACE data and averaged SD in grey matter (GM) mask are reported as the quality of fMRI dataset.
Time-series of temporal derivative of variance over voxels (DV) (10) in GM and framewise displacement (FD) (11) were calculated. In addition to DV and FD, intra volume motion index, the derivatives of intra-volume volume displacement (iVD) and z component of iVD (iVDz) with the temporal derivative 6 motion parameters at slice-wise time points from SLOMOCO. Time-series of variance over voxels based on the averaged motion corrected images (DVVOL/SLOMOCO) is also calculated, presenting the temporal quality of motion corrected data. Temporal DV, FD, and iVD(z) values were compared to DVVOL/SLOMOCO. Five volumes with top 5 largest FD, DV, iVD(z) value were removed and averaged SD value in GM with each motion censored SIMPACE data were calculated.
Two different motion correction and pipelines were used. SIMPACE data was 3d rigid volume motion corrected (VOLMOCO). After VOLMOCO, 6 rigid motion parameters were regressed out with voxelwise partial volume (PV) nuisance regressors (8). SLOMOCO was applied with 12 volume-/slice-wise motion nuisance and PV regressor (6, 9). We calculate temporal standard deviation (SD) of the residual signal in VOL/SLOMOCO SIMPACE data and averaged SD in grey matter (GM) mask are reported as the quality of fMRI dataset.
Time-series of temporal derivative of variance over voxels (DV) (10) in GM and framewise displacement (FD) (11) were calculated. In addition to DV and FD, intra volume motion index, the derivatives of intra-volume volume displacement (iVD) and z component of iVD (iVDz) with the temporal derivative 6 motion parameters at slice-wise time points from SLOMOCO. Time-series of variance over voxels based on the averaged motion corrected images (DVVOL/SLOMOCO) is also calculated, presenting the temporal quality of motion corrected data. Temporal DV, FD, and iVD(z) values were compared to DVVOL/SLOMOCO. Five volumes with top 5 largest FD, DV, iVD(z) value were removed and averaged SD value in GM with each motion censored SIMPACE data were calculated.
Results:
Figure 1 shows the temporal characteristic of 10 motion corrected SIMPACE data after VOLMOCO and SLOMOCO and corresponding motion indexes. FD, DV, iVD and iVDz show a high correlation coefficient of 0.32 (p < 10-4), 0.25 (p < 0.01), 0.36 (p < 10-5) and 0.48 (p < 10-9) with DVVOLMOCO, respectively, but not showing significant correlation with DVSLOMOCO. It should be noted that DVSLOMOCO is smaller with less spontaneous peak than DVVOLMOCO.
Figure 2 shows average SD in GM in VOLMOCO- and SLOMOCO-corrected SIMPACE data with/out censoring 5 volumes. Also shown (horizontal dashed line) is the SD in a SIMPACE acquisition with no motion injected, which can be considered the baseline, or truth. SD in GM is reduced after censoring when using VOLMOCO, but not with SLOMOCO.
Figure 2 shows average SD in GM in VOLMOCO- and SLOMOCO-corrected SIMPACE data with/out censoring 5 volumes. Also shown (horizontal dashed line) is the SD in a SIMPACE acquisition with no motion injected, which can be considered the baseline, or truth. SD in GM is reduced after censoring when using VOLMOCO, but not with SLOMOCO.
·Figure 1. Temporal characteristic of 10 SIMPACE data and different motion indexes. X axes represent time unit (volume number) and Y axes are arbitrary units, but same scaled across SIMPACE datasets.
·Figure 2. Averaged standard deviation (SD) in grey matter (GM) after VOLMOCO and SLOMOCO in 10 SIMPACE data. * indicates the significant difference from no censoring data (A, p < 0.05).
Conclusions:
We find that time-series FD, DV and iVD(z) is highly correlated to the residual motion artifact in VOLMOCO SIMPACE data, but not in SLOMOCO, indicating that there is more residual motion artifact when using VOLMOCO as previously presented (12). Censoring using FD, DV and iVD(z) indexes reduce the residual motion artifact in VOLMOCO-corrected SIMPACE data, but not in SLOMOCO. This result suggests that SLOMOCO-corrected data has minimal levels of residual motion artifact compared to VOLMOCO-corrected data. Average SD in GM for SLOMOCO-corrected data is 5.45±0.84 without censoring. Considering that the baseline SD (i.e. no motion injected) is 4.92 in GM, SLOMOCO output removes the residual artifact from spontaneous inter-/intra-volume motion effectively. Future studies with larger samples of realistic motion simulated data are planned to further validate these results.
Modeling and Analysis Methods:
Motion Correction and Preprocessing 1
Novel Imaging Acquisition Methods:
BOLD fMRI 2
Keywords:
FUNCTIONAL MRI
Other - head motion
1|2Indicates the priority used for review
Provide references using author date format
2. A. Jiang et al., Motion detection and correction in functional MR imaging. Human Brain Mapping 3, 224-235 (1995).
3. K. J. Friston, S. Williams, R. Howard, R. S. Frackowiak, R. Turner, Movement-related effects in fMRI time-series. Magn Reson Med 35, 346-355 (1996).
4. S. Grootoonk et al., Characterization and correction of interpolation effects in the realignment of fMRI time series. Neuroimage 11, 49-57 (2000).
5. S. Thesen, O. Heid, E. Mueller, L. R. Schad, Prospective acquisition correction for head motion with image-based tracking for real-time fMRI. Magn Reson Med 44, 457-465 (2000).
6. E. B. Beall, M. J. Lowe, SimPACE: generating simulated motion corrupted BOLD data with synthetic-navigated acquisition for the development and evaluation of SLOMOCO: a new, highly effective slicewise motion correction. Neuroimage 101, 21-34 (2014).
7. S. Kim, K. Sakaie, I. Blumcke, S. Jones, M. J. Lowe, Whole-brain, ultra-high spatial resolution ex vivo MRI with off-the-shelf components. Magn Reson Imaging 76, 39-48 (2021).
8. W. Shin, M. J. Lowe. Effective removal of the residual head motion artifact after motion correction in fMRI data. in 2023 ISMRM & ISMRT Annual Meeting & Exhibition, Toronto, #1821. (2023)
9. W. Shin, M. J. Lowe. The development of Inter-/intra-volume motion correction algorithm for fMRI using a custom MRI acquisition with prospectively injected motion. in 2023 ISMRM & ISMRT Annual Meeting & Exhibition, Toronto, #2001. (2023)
10. C. D. Smyser et al., Longitudinal analysis of neural network development in preterm infants. Cereb Cortex 20, 2852-2862 (2010).
11. J. D. Power, K. A. Barnes, A. Z. Snyder, B. L. Schlaggar, S. E. Petersen, Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion. NeuroImage 59, 2142-2154 (2012).
12. J. D. Power et al., Methods to detect, characterize, and remove motion artifact in resting state fMRI. Neuroimage 84, 320-341 (2014).