github Deep-MI/FastSurfer v2.4.0

one day ago

FastSurfer v2.4.0

This version is a big one and introduces some really exciting new functionality! We almost made it a major release, but since it should not really break any compatibility, we will keep some more really cool things in stock for you for v3. Hard to believe we can do even more than what we have in this release:

  • Longitudinal Stream
    You have waited long for the longitudinal pipeline in FastSurfer and now it is finally here. In our tests we see increased reliability and sensitivity of this new stream, but not only with respect to FastSurfer's independent processing, but even with respect to our older FreeSurfer longitudinal pipeline. We are really excited about this new stream. There is no reason to wait, try for yourself and see Long-Documentation to get started.

  • Edits
    We have added a few ways to interact with FastSurfer and introduce manual edits, such as white matter and pial surface edits. These should rarely be necessary, but if you want to intervene, now you can. Find all the details in our Edits-Documentation.

  • Parallel

    • We have cleaned up the many ways that things run in parallel. This makes the interface much simpler for run_fastsurfer.sh: you only need to use --threads to select a target number of cores now (or --threads_seg and --threads_surf for more control over the threads in each block). The --parallel argument is deprecated and run_fastsurfer.sh will automatically process both hemispheres in parallel, if you use more than 1 thread (1 is currently still the default). Note, we no longer double the number of threads used during parallel surface processing. So for threads=2, each hemisphere gets one thread, the rest of the code uses 2 threads.
    • Furthermore, we added flags to brun_fastsurfer.sh to allow specifying the number of parallel cases in the segmentation block and the surface block separately, for a more fine grained control (--parallel_seg and --parallel_surf), e.g. of GPU usage (in seg) vs. CPU (in surf) and even using multipe GPUs.
    • Our new long_fastsurfer.sh now can make use of this functionality when you want to run multiple time points in parallel, try --parallel_surf there.
  • Other Changes

    • We cleaned up the code and introduced code style checks.
    • Documentation has been updated https://deep-mi.org/FastSurfer/dev/index.html
    • Improved logging
    • Bug fix to HypVINN for large images and other minor fixes
    • Bumped versions of dependencies

Notice

All the amazing functionality of FastSurfer is brought to you for free. You can help us, by making sure you cite our papers in you publications, not only the initial FastSurfer paper or the VINN for high-res support, but also the papers describing the individual modules that you use (Cerebnet, Hypvinn, or our older papers on Longitudinal Processing in FreeSurfer which are the basis for the new longitudinal pipeline).

Future Highlights

Our roadmap includes many things, some of them are listed here:

  • Lesion Inpainting to process cases with lesions, tumor, resection cavities or other pathologies (this already works with our Lesion Inpainting Tool
  • Improved Corpus Callosum and Fornix segmentation, including AC and PC
  • Improved eTIV estimation via segmentation
  • Addition of more labels, e.g. for WM sub-segmentation, skull and bone segmentation
  • FastPlex for the accurate Choroid Plexus segmentation
  • T2 / Flair segmentation

Detailed Changes (selection)

Full Changelog: v2.3.3...v2.4.0

Don't miss a new FastSurfer release

NewReleases is sending notifications on new releases.