FreeSurfer / FastSurfer

FreeSurfer and FastSurfer are the two most widely used open-source tools for extracting structural measurements from T1-weighted MRI scans. They are the entry point for almost every structural brain age pipeline.

FreeSurfer

Developed at the Martinos Center at Massachusetts General Hospital, FreeSurfer has been the field standard since the early 2000s. Its core pipeline (recon-all) takes a raw T1w scan and performs:

1. Skull stripping — removes non-brain tissue
2. Intensity normalisation — corrects for field inhomogeneity
3. Tissue segmentation — separates grey matter, white matter, and CSF
4. Surface reconstruction — models the cortex as a 2D sheet, capturing sulcal folding
5. Parcellation — assigns each vertex or voxel to an anatomical region (using atlases such as Desikan–Killiany or Destrieux)
6. Feature extraction — outputs thickness, volume, surface area, and curvature for ~70–150 regions per hemisphere

The full recon-all pipeline takes 4–8 hours per scan on a standard CPU.

FastSurfer

FastSurfer, developed at the German Center for Neurodegenerative Diseases, replicates much of FreeSurfer's output using deep learning. Its key advantages:

• Whole-brain segmentation in ~1 minute on a GPU (versus hours for FreeSurfer)
• Comparable accuracy to FreeSurfer on standard datasets
• Better performance on non-standard or lower-quality scans in some benchmarks

FastSurfer is increasingly used in large-scale studies and clinical pipelines where throughput matters.

What the outputs look like

Both tools produce a table of regional measurements. A typical entry might read:

Left entorhinal cortex — mean thickness: 3.41 mm, volume: 1 842 mm³

These measurements form the feature vector that brain age models use as input. A model trained on tens of thousands of healthy participants learns which combinations of regional thinning and volume loss are typical for a given age — and flags deviations.

Caveats

• FreeSurfer and FastSurfer outputs are not perfectly interchangeable; a brain age model trained on one should ideally not be applied to output from the other without validation.
• Both pipelines can fail or produce artefacts on low-quality scans (motion, unusual pathology, very elderly brains with extreme atrophy).
• Version changes between FreeSurfer releases can introduce systematic shifts in measurements — a known challenge for longitudinal studies.

Read more

• T1-weighted MRI
• Cortical thickness
• How we estimate brain age