deepmi/fastsurfer repository overview

⁠FastSurfer

FastSurfer is a fast and accurate deep-learning pipeline for the analysis of human brain MRI. It is a fully compatible FreeSurfer alternative for volumetric and surface-based thickness analysis, supports sub-millimeter resolutions down to 0.7 mm, and sub-segments neuroanatomical structures such as the corpus callosum, cerebellum and hypothalamus. The whole brain segmentation takes a few minutes on a GPU, the surface reconstruction 20 to 40 minutes.

📖 Documentation: https://deep-mi.org/FastSurfer/stable/⁠

⁠Getting started

The documentation has commands for the current release, ready to copy:

• Linux: Apptainer/Singularity or Docker⁠
• Windows: Docker in WSL2⁠
• macOS⁠: on Apple silicon, use the native macOS package, so you don't need these images.

⁠Tags

Which image fits your GPU⁠ lists the supported GPUs and the minimum NVIDIA driver of each CUDA image. If FastSurfer cannot use your GPU, it tells you which image to use instead. For reproducible results, use a versioned tag rather than latest.

⁠FreeSurfer license

The surface reconstruction uses FreeSurfer tools, which the images include (stripped down to what FastSurfer needs). FreeSurfer's license file is free but not included: register here⁠ and pass it to FastSurfer with --fs_license. The segmentation alone (--seg_only) does not need a license.

⁠License and citation

FastSurfer is released under the Apache License 2.0⁠; the included FreeSurfer tools are subject to the FreeSurfer license⁠. If you use FastSurfer in research publications, please cite the papers listed in the documentation⁠.

Source code: https://github.com/Deep-MI/FastSurfer⁠