Whole-Brain Segmentation (FastSurferVINN)

The core module of FastSurfer (asegdkt): a deep learning network (FastSurferVINN) that segments the whole brain into 95 cortical and subcortical structures, following FreeSurfer’s DKT atlas, and computes volume statistics for them. All other modules build on its result.

What it computes

  • A segmentation of cortical and subcortical structures (aparc.DKTatlas+aseg.deep.mgz), equivalent to FreeSurfer’s aparc.DKTatlas+aseg.mgz, and a simplified subcortical segmentation (aseg.auto_noCCseg.mgz).

  • A brain mask and a bias-field corrected image.

  • Volume statistics for all structures, corrected for partial volume effects (aseg+DKT.stats).

  • Optionally, with --tal_reg, the estimated total intracranial volume (eTIV) in the statistics.

The output files are listed in the output files overview. If the surface reconstruction also runs, it refines the segmentation along the cortex with the surfaces, and its updated segmentations and statistics are the ones to use.

What it needs

  • A T1-weighted image, see the requirements to input images.

  • Voxel sizes between 0.7 mm and 1 mm are supported natively (smaller voxels are experimental), see --vox_size.

  • No FreeSurfer license, unless you switch on the optional Talairach registration (--tal_reg), which is only used to estimate the total intracranial volume (eTIV); that needs a FreeSurfer license.

The segmentation takes a few minutes on a GPU, and longer on the CPU.

Options

The module runs by default. The most important options of run_fastsurfer.sh for it:

  • --seg_only: run the segmentation modules only, without the surface reconstruction.

  • --no_asegdkt: skip this module, for example to rerun only other modules on an existing segmentation.

  • --asegdkt_segfile <path>: where to write the segmentation (default mri/aparc.DKTatlas+aseg.deep.mgz).

  • --vox_size <0.7-1|min>: the voxel size to process at (default min, from the input image).

  • --no_biasfield: skip the bias-field correction and the partial-volume corrected statistics.

  • --tal_reg (with --seg_only, optional): add the Talairach registration, only used to estimate eTIV; it needs a FreeSurfer license, and --3T uses the 3T atlas for it. The surface reconstruction always computes eTIV.

  • --keepgeom (with --seg_only): write the outputs in the geometry of the input image.

Usage:

run_fastsurfer.sh --seg_only --sd <subjects_dir> --sid <subject_id> --t1 <t1_path>

All options are described in the run_fastsurfer.sh reference. To run the network script directly, see FastSurferVINN in the command reference.

References

If you use FastSurfer in your research, please cite:

  • Henschel L, Conjeti S, Estrada S, Diers K, Fischl B, Reuter M. FastSurfer - A fast and accurate deep learning based neuroimaging pipeline. NeuroImage 219 (2020), 117012. doi:10.1016/j.neuroimage.2020.117012

  • Henschel L*, Kuegler D*, Reuter M. (*co-first) FastSurferVINN: Building Resolution-Independence into Deep Learning Segmentation Methods - A Solution for HighRes Brain MRI. NeuroImage 251 (2022), 118933. doi:10.1016/j.neuroimage.2022.118933