Motion correction and scrubbing for task-based fMRI

1

Hi all,

I have a question regarding motion correction for task-based fMRI data. Unfortunately my data contains a bit of motion due to the population as well as task performance inside the MRI scanner.

I have preprocessed my data using fMRIPrep (version 23.2.0), and ICA-AROMA is no longer recommended to add to the fMRIPrep pipeline.
However, I was wondering if it is recommended to use ICA-AROMA after preprocessing finished and after spatial smoothing, before high-pass filtering and before nuisance regression? I’ve read multiple papers and some suggest ICA-AROMA is recommended, while others disagree.
I’m relatively new to neuroimaging analyses, so I would appreciate your input.

In addition, I was thinking of excluding participants with excessive motion after applying all relevant correction methods. Do you have any references for exclusion criteria for task-based fMRI?

Thank you!

2

Hi @Joni, and welcome to Neurostars!

Among nuisance regression, aCompCor is an efficient way to denoise your fMRI data. Regarding motion, you can use additional motion regressors than the 3 rotations and 3 translation, i.e the derivatives of those (12 motion regressors in total) or even the square of those 12 (24 motion regressors in total).

Relevant paper for aCompCor:
Behzadi, Y. A Component Based Noise Correction Method (CompCor) for BOLD and Perfusion Based fMRI. 2007, 12.

Interesting paper about resting state and task fMRI advanced denoising pipelines:
De Blasi, B.; Caciagli, L.; Storti, S. F.; Galovic, M.; Koepp, M.; Menegaz, G.; Barnes, A.; Galazzo, I. B. Noise Removal in Resting-State and Task fMRI: Functional Connectivity and Activation Maps. J. Neural Eng. 2020, 17(4), 046040. ShieldSquare Captcha.

About exclusion criteria:

Time points with FD greater than 0.9 mm are censored".

  • In this study:
    Silesset al. Image Acquisition and Quality Assurance in the Boston Adolescent Neuroimaging of Depression and Anxiety Study. NeuroImage: Clinical 2020, 26, 102242. Redirecting.

Fig. 14 shows the average percentage of volumes with FD > 0.9 mm, which has been used previously as a threshold for discarding fMRI volumes in adolescents (Siegel et al., 2014). Typically, when more than 20% of volumes surpass this threshold, the subject’s scan is discarded from further analyses.

Note that the TR of your functional scan may be taken into account for the exclusion criteria. This is a personal observation, but I think that depending on how long is your TR, the inter-volume motion parameters estimated from classic Head motion correction techniques may be different resulting in framewise displacement values depending on the TR.

Also these papers are interesting in taking into account breathing effect in the motion estimation in fMRI:

Gratton et al. Removal of High Frequency Contamination from Motion Estimates in Single-Band fMRI Saves Data without Biasing Functional Connectivity. NeuroImage 2020, 217, 116866. Redirecting.

Fair et al. Correction of Respiratory Artifacts in MRI Head Motion Estimates. NeuroImage 2020, 208, 116400. Redirecting.

2 Likes
3

Thank you for the great resources @jsein ! I wonder whether you (or anyone else) have thoughts on the following concerning these papers:

It appears the censoring procedures rely on Siegel (2014). If I understand the Siegel paper correctly, it actually indicates that censoring is superior to motion regressors, even with better performance for “censoring only” compared to “censoring plus motion regressors”. Interestingly, with uncensored data, only the derivatives of the motion regressors appeared to improve the signal, not the six basic motion regressors. This could be interpreted as there being no need to actually include motion regressors and, if one still wants to include motion regressors, only derivatives. Moreover, they find that FD > .90 is a good cutoff for censoring, but fmriprep uses a much more stringent cutoff of FD > .50. This made me wonder why people use this more stringent fmriprep cutoff and still include motion parameters.

I also tried to look for more recent sources, but only found studies on resting state fMRI. In the study by Parkes et al. (2018), the different control parameter choices are unfortunately a bit confounded, but it still looks to me like motion parameters are not doing much (generally, it looks to me like global signal regression is doing most of the lifting here).

Siegel et al. (2014) Statistical improvements in functional magnetic resonance imaging analyses produced by censoring high‐motion data points - PMC

Parkes et al. (2018) An evaluation of the efficacy, reliability, and sensitivity of motion correction strategies for resting-state functional MRI - ScienceDirect

1 Like