We are conducting a study on children that are imaged around 11 years of age. In the index group we can expect that the children cannot perform/cope with the MRI as well as in the control group (e.g. we expect that they may move more etc). However, we are, and have to be, blinded in the analysis. We have collected multishell DWI data (30 dir with b1000 and 60 dir with b2500 in two consecutive scans) and an additional dataset of reversed gradient polarity (with 6 dir at b200). Quality of DWI in general good with adequate SNR. We are pre-processing with topup+eddy. This works fine.

However, there are artefacts in the raw data. Mainly signal dropout artifacts due to pulsation or head motion. Primarily affecting the posterior fossa (pulsation) but also in some slices scattered among the volumes in the superior fossa (head motion here). The larger signal-dropouts seems to remain after eddy so at least this procedure is not blending them into the eddy-corrected data.

Have you any advice on how to deal such artefacts in DWI? How to detect corrupted slices/volumes? Is it best to exclude certain volumes before fitting a diffusion model? This would mean dropping a lot of valuable data due artefacts in a few slices. Alternatives would be to use a post-processing check like RESTORE or HOMOR (are there other suitable for FSL's diffusion model??) to detect outlier voxels, or better (?) a pre-processing implementation to combine any of them with DROP-R (Morris et al, MRM 2011)??

Advice appreciated!

Cheers,

Finn