Summary of what happened:
Dear all,
I am currently struggling with the brain masks generated in fMRIPREP run in singularity for some pilot data.
The best results I am getting currently is using the LTS Version, but even here the brain mask is not including the whole brain.
This has happened for several runs / participants using the sequence, here are just a few examples of a particular run using different outputs of fMRIPREP.
I think what is irritating me the most is that MRIQC and BOLD masks calculated by hand (e.g. nilearn) are more accurate than the one by fMIRPREP.
Happy if you have any suggestions what could help here 
Screenshot 1:
This is the best resulting brain mask I could achieve so far:
fMRIPrep version: 20.2.7
fMRIPrep command: /usr/local/miniconda/bin/fmriprep /projectFolder/raw /projectFolder/derivatives/fmriprep-20.2.7_syn participant --participant-label 03 -w /work --output-spaces MNI152NLin2009cAsym:res-native anat --random-seed 1022 --ignore fieldmaps --use-syn-sdc --force-syn --nprocs 4 --mem_mb 20000 --no-submm-recon --notrack --fs-license-file /projectFolder/code/fmriprep/license.txt
Screenshot 2:
Using newer versions of fMRIPREP the results become worse.
fMRIPrep version: 23.0.2
fMRIPrep command: /opt/conda/bin/fmriprep /projectFolder/raw /projectFolder/derivatives/fmriprep-23.0.2_2 participant --participant-label 03 -w /work --output-spaces MNI152NLin2009cAsym:res-native anat --random-seed 1022 --no-submm-recon -v --skip_bids_validation --nprocs 4 --mem_mb 20000 --notrack --fs-license-file /projectFolder/code/fmriprep/license.txt
Screenshot 3:
In some runs or using syn-sdc, resulting brain masks came out like this.
fMRIPrep version: 23.0.2
fMRIPrep command: /opt/conda/bin/fmriprep /projectFolder/raw /projectFolder/derivatives/fmriprep-23.0.2_syn participant --participant-label 03 -w /work --output-spaces MNI152NLin2009cAsym:res-native anat --random-seed 1022 --skip_bids_validation --ignore fieldmaps --clean-workdir --nprocs 4 --mem_mb 20000 --force-syn --use-syn-sdc --no-submm-recon --notrack --fs-license-file /projectFolder/code/fmriprep/license.txt
Aside: I had similar brain masks in older data we acquired, where results of an older version of fMRIPREP seemed to be fine.
MRIQC:
Sequence details
Data has been acquired on a PRISMA 3T scanner, with a TR of 1.791, voxel resolution 1.4 x 1.4 x 1.4 mm and 20 slices, we used a 6/8 partial fourier, PhaseEncdoingDirection j- / AP
The particular sequence shown here lasted for ~24 mins.
For susceptibility correction (if applied in the examples above) we acquired a few epis with reverse phase encoding.
Thanks!