Correction of respiratory artifacts in MRI head motion estimates

Abstract

Head motion represents one of the greatest technical obstacles for brain MRI. Accurate detection of artifacts induced by head motion requires precise estimation of movement.However, this estimation may be corrupted by factitious effects owing to main field fluctuations generated by body motion. In the current report, we examine head motion estimation in multiband resting state functional connectivity MRI (rs-fcMRI) data from the Adolescent Brain and Cognitive Development (ABCD) Study and a comparison 'single-shot' dataset from OregonHealth & Science University. We show unequivocally that respirations contaminate movement estimates in functional MRI and that respiration generates apparent head motion not associated with degraded quality of functional MRI. We have developed a novel approach using a bandstop filter that accurately removes these respiratory effects. Subsequently, we demonstrate that utilizing this filter improves post-processing data quality. Lastly, we demonstrate the real-time implementation of motion estimate filtering in our FIRMM (Framewise Integrated Real-Time MRI Monitoring) software package.(FIRMM) . FIRMM provides near instantaneous analysis of head motion (framewise displacement = FD) during scanning. This feature enables the duration of resting state fMRI scanning to be dynamically determined, that is, for as long as is necessary to acquire a prescribed quantity of data meeting a fixed quality assurance criterion. Thus, FIRMM helps to ensure that an adequate quantity of resting state data is acquired in most participants.It also reduces the need for 'overscanning' all participants in order to ensure adequate acquisitions across subjects. Moreover, subjects who cannot suppress excessive motion can be identified promptly and efficiently excluded from the study. Additionally, FIRMM alerts the scanner operator to changes in participant behavior (e.g., sleep, or increased movement because of discomfort), and also can be utilized to give motion feedback to participants themselves. These advances have been thoroughly detailed in prior reports Greene et al., 2018).Here, we address new issues that have arisen when calculating head motion estimates (FD) fMRI data. Multiband imaging (simultaneous multi-slice sequences [SMS]) is a recently developed technique that substantially speeds up fMRI by exciting multiple slices simultaneously (Feinberg and Yacoub, 2012;Moeller et al., 2010;Uğurbil et al., 2013;Xu et al., 2013). Thus, the time needed to acquire one whole brain volume can be reduced to well below 1 sec. While multiband sequences offer several clear advantages and are now becoming widely used, they come at the cost of reducing the temporal signal to noise ratio (tSNR; (Chen et al., 2015)) and, potentially, new types of 'slice-leakage' artifacts (Barth et al., 2016;Todd et al., 2016).Another unanticipated consequence of the improved temporal resolution of SMS sequences is corruption of head motion estimation (FD) arising from the interaction between echo-planar imaging (EPI) and small perturbatio...

Dynamic time warping outperforms Pearson correlation in detecting atypical functional connectivity in autism spectrum disorders

“…The Artifact Detection Toolbox (ART, as installed with conn v17f) was used to identify outliers in the functional image time series from the resulting 6 motion parameters (3 translational and 3 rotational) that had frame-wise displacement (FD) >0.5 mm and/or changes in signal intensity that were greater than three standard deviations in the child/adolescent cohort. As oscillations due to respiration are prominent in motion parameters derived from multiband EPI realignment ( Fair et al, 2018 ) and would result in unnecessary censoring of large segments of data in some participants, the thresholds to detect outliers in the adolescent and adult cohorts were more lenient than those used for standard resting state fMRI acquisitions with slower TRs (FD>0.9 mm and/or changes in signal intensity that were greater than five standard deviations). In order to ensure that none of our findings were due to differences in apparent motion between the ASD and TC groups, groups in each cohort were matched on RMSD (see Table 1 ) calculated from rigid-body realignment (of the raw data prior to TOPUP correction in case of the adolescent and adult cohort) parameters, and partial correlations controlling for RMSD were used when assessing brain-behavior relationships (Analysis 5).…”

On the analysis of rapidly sampled fMRI data

“…Here, we review a few caveats that warrant particular attention when removing nuisance factors from data collected with short TRs. First, respiratory artifacts, which are less noticeable at conventional 2~3 s TRs due to aliasing effects, become pronounced in head motion estimates as TR shortens (Fair et al, 2018;Siegel et al, 2017), as illustrated in Fig. 1 (a) and supplementary material SM2).…”

“…Therefore in this case, deleterious respiratory fluctuations may be introduced into fMRI datasets after regression against the estimated motion traces. To avoid this deleterious effect, one can either notch-filter the quasi-periodic respiration waveforms from motion estimates prior to nuisance regression (Fair et al, 2018), or include co-varying physiological parameters (e.g., RETROICOR covariates (Glover et al, 2000)) together with the estimated motion traces in the nuisance regression.…”

Impaired motor skills and atypical functional connectivity of the sensorimotor system in 40- to 65-year-old adults with autism spectrum disorders

“…The Artifact Detection Toolbox (ART, as installed with conn v17f) was used to identify outliers in the functional image time series from the resulting 6 motion parameters (3 translational and 3 rotational) that had frame-wise displacement (FD) >0.9mm and/or changes in signal intensity that were greater than five standard deviations. As oscillations due to respiration are prominent in motion parameters derived from multiband EPI realignment (Fair et al, 2018) and would result in unnecessary censoring of large chunks of data in some participants, the thresholds to detect outliers were more lenient than those used for standard resting state fMRI acquisitions with slower TRs (but see Supplementary Methods and Figure S3 for supplementary analyses with stricter censoring thresholds). In order to ensure that none of our findings were due to differences in apparent motion between groups, participants were matched on RMSD calculated from rigid-body realignment of the raw data prior to TOPUP correction (Table 1).…”