--

Juan Eugenio Iglesias

http://www.jeiglesias.com

Hello everyone, I am using SynthSR (FreeSurfer 8.1.0, mri_synthsr) to standardize T1-weighted MRI scans across a multi-protocol clinical dataset. The majority of subjects process correctly; however, for a small subset of scans, the SynthSR output contains prominent white rectangular/block artefacts over the brain volume.

Key observations:

• The artefacts are not present in the original T1 images

• They appear only in the SynthSR output

• The affected original scans are already 1 × 1 × 1 mm isotropic

• Original images otherwise appear visually normal (no obvious truncation or corruption)

• Artefacts look axis-aligned (rectangular slabs), suggesting a reconstruction failure rather than anatomy

Example workflow:

mri_synthsr \
  --i sub-XX_T1w.nii \
  --o sub-XX_synthsr.nii.gz

Questions:

1. Are these white block artefacts a known failure mode of SynthSR for certain inputs?

2. Are there recommended preprocessing checks (e.g., header/orientation fixes) to reduce this issue?

3. Would trying the --v1 model be advisable in such cases?

4. If the original scan is already clean and isotropic, is it methodologically acceptable to exclude SynthSR output and retain the original image for downstream processing?

I am currently treating SynthSR as an exploratory preprocessing step and excluding affected subjects from the SynthSR-based pipeline, but I would appreciate guidance on best practices.

Thank you for any insights