Dear Kwangyeol,
Looking at the intensity distribution of the (I guess smoothed) EPI with decreasing signal towards medial and ventral regions it's likely receive coil heterogeneity or "flare" of a multi-channel coil. See http://practicalfmri.blogspot.de/2012/04/common-persistent-epi-artifacts-receive.html for more information. You can avoid this at the scanner with a corresponding correction/calibration before the actual sequence, the names differ depending on manufacturer, PURE for GE, Prescan Normalize for Siemens, CLEAR for Philips. It should be possible to save both corrected and uncorrected data, but make sure that you don't accidentally combine this correction with another (I vaguely recall someone somewhen talking about motion correction being enabled, resulting in coil heterogeneity corrected, motion corrected data and "uncorrected" motion corrected data). If you don't account for intensity differences then you will indeed lose those regions with lower intensities, as SPM uses a rather funny procedure to restrict the model to voxels above a certain intensity only (which usually works well as long as there's not much intensity difference, but can fail with e.g. multi-channel data).
For now, as you seem to have acquired some data already, the simplest solution would be to lower the necessary masking threshold (within SPM12 you can do so when setting up the model, in older versions, you will have to edit defaults.mask.thresh in spm_defaults.m).
However, note that motion correction should benefit from coil heterogeneity corrected data. Thus I would usually propose something like this in case you haven't corrected the data at the scanner:
- copy your raw EPI series
- realign and reslice a mean image
- forward the mean into Segment, save a bias-corrected version and a bias field
- apply the corresponding bias field onto the raw (unrealigned!) EPI series via Imcalc, dunno right now, either you have to multiply the EPIs with the bias field or to divide (the output should have a more homogeneous intensity across the brain, possibly you should try out different bias field correction settings to get an impression what works best with your data)
- start with your usual preprocessing pipeline
This might likely have to be adapted to your data, as you should have three series acquired with different TEs, and I'm not sure whether one would create one bias field for each of the series and apply these within series, or whether you would want to average across the three bias fields and apply this averaged version onto the three series.
But I would be surprised if coil heterogeneity hadn't been considered for the MESMS protocol. Maybe you already have a corresponding coil profile file which can be used for corrections? Thus I would suggest to contact the MESMS/MEICA authors, they should know best how to properly correct the data. Maybe you can tell us about their suggestions then, considering those who might run into similar issues and browse the list. Thank you!
Best
Helmut
|