MRI-to-FEM Pipelines for FEniCS(x)

This document outlines the workflows used at Simula for converting raw MRI data into finite element meshes suitable for FEniCS/FEniCSx simulations. It covers current software choices, specific user pipelines, dataset locations, and future development needs.

1. Quick Reference: Current Workflows

A summary of who is using what, serving as a point of contact for specific variations of the pipeline.

Contact	Meshing Engine	Segmentation Source	Dataset / Repo	Key Notes
Ingvild	Ftetwild	FreeSurfer / SynthSeg	OpenNeuro ds004478	Uses mri2mesh. Feature request: Node index tracking.
Andreas	SVMTK	FreeSurfer	SVMTK Repo	Standard SVMTK pipeline.
Henrik	Ftetwild	NeuroQuant	Biophysics Repo	Uses a config-file driven approach. Handles NeuroQuant labels (different from FS).
Cecile	Ftetwild	FreeSurfer / SynthSeg	CerebroSpinalFlow	Based on Marius Causemann's transport repo.
Timo / Jørgen	SVMTK / PyVista	FreeSurfer	Gonzo	SVMTK preferred for surface preprocessing (keeps sulci better than PyVista).

---

2. Pipeline Stages

Stage 1: Raw Data to Images

Conversion of scanner output (typically DICOM) to research-standard formats (NIfTI).

• Tools: dcm2niix (Standard), proprietary scanner tools.
• Target Modalities:
  • T1/T2: Structural anatomy (geometry).
  • Look-Locker: T1 mapping.
  • DTI: Diffusion Tensor Imaging (anisotropy tensors).
  • fMRI/BOLD: Functional activation maps.
  • FLAIR: Fluid-attenuated inversion recovery.

Stage 2: Registration

Aligning different image modalities (e.g., registering DTI to the T1 structural scan) or registering subjects to a common atlas (MNI).

• Greedy: Fast, effective for standard rigid/affine registration.
• ANTs: Gold standard for non-linear warping and complex registration.
• FSL FLIRT: Robust linear registration.

Stage 3: Segmentation

Classifying voxels into tissue types (White Matter, Gray Matter, CSF).

A. FreeSurfer (Standard)

• Command: recon-all
• Pros: Industry standard, highly detailed cortical surfaces, separates hemispheres.
• Cons: Slow (hours), requires high-quality T1.
• Output: aseg.mgz, lh.pial, rh.white.

B. Deep Learning (SynthSeg)

• Command: mri_synthseg and recon-all-clinical
• Pros: Fast (minutes), robust to domain shifts and lower resolution scans.
• Cons: Surfaces may lack the topological guarantees of FreeSurfer.
• Note: Also part of FreeSurfer.

Other alternatives we migth want to consider

• FastSurfer
• NextBrain
• GOUHFI
• SAM3
• NeuroQuant - This is a commercial tool used by Anders Dale

Stage 4: Surface Extraction

Converting voxel labels into triangular surface meshes (STL/PLY).

• SVMTK:
  • Works well with output from Freesurfer. Provide fixes for issues introduced by Freesurfer
• FreeSurfer:
  • Native extraction via mri_tessellate or mri_mc.
• PyVista:
  • Uses VTK marching cubes. Good for visualization, but requires careful smoothing to avoid non-manifold artifacts in complex brain geometry.

Resources

• cdaversin/CerebroSpinalFlow
• mri2mesh
• MariusCausemann/brain-PVS-SAS-transport

Stage 5: Meshing (Surface $\to$ Volume)

Creating the tetrahedral mesh for FEniCS / FEniCSx.

1. Wildmeshing (Ftetwild)

• Implementations:
  • wildmeshing-python (Wrapper by Cecile or Marius).
  • pytetwild (PyVista wrapper).
  • C++ Source.
• Pros: Extremely robust. Can mesh "dirty," self-intersecting, or non-manifold surfaces where other algorithms fail.
• Cons: Tricky to install. It is possible to install with pip on linux x84 and mac arm (but not linux on arm, i.e docker on arm mac).
• Note: They are also working on a new version, but this is not ready for us to use.
• Examples:
  • Tutorial: https://scientificcomputing.github.io/fenics-in-the-wild
  • https://github.com/MariusCausemann/brain-PVS-SAS-transport

2. SVMTK (Surface Volume Meshing Toolkit)

• Implementation: GitHub Link
• Pros:
  • Wraps CGAL.
  • Specifically designed for biomedical domains.
  • Excellent utilities for surface repair and subdomain marking.
• Cons:
  • Only maintained by Lars Magnus Valnes
  • Tricky to install (conda is the best alternative)
• Examples:
  • https://github.com/jorgenriseth/gonzo

Stage 6: Data Mapping (FEniCSx)

Mapping physiological data from NIfTI images to the generated mesh.

• Workflow:
  1. Load NIfTI (via nibabel).
  2. Align coordinate systems (RAS vs LPS).
  3. Create FEniCS FunctionSpace (e.g., DG0 for voxels, CG1 for continuous fields).
  4. Probe/Interpolate values.
• Applications:
  • Tissue properties (conductivity, elasticity).
  • DTI-based diffusion tensors.
  • Concentration maps.
• Examples:
  • https://github.com/jorgenriseth/gonzo
  • Utility functions in scifem

---

3. Example Recipe: "The Marius Pipeline"

A hypothetical workflow to go from raw scan to a Stokes + Advection-Diffusion simulation.

1. Preprocessing:
   • Convert DICOM to NIfTI (dcm2niix).
2. Segmentation:
   • Run mri_synthseg --i input.nii --o seg.nii (Fast, robust).
3. Surface Extraction (PyVista):
   • Extract isosurfaces for GM, WM, CSF and perhaps ventricles. Smooth surfaces with PyVista.
   • Save output to .ply files.
4. Meshing (Ftetwild):
   • Generate tetrahedral mesh from the cleaned surfaces using boolean (union) operations.
   • Mark subdomains (Cell tags) and boundaries (Facet tags).
5. Simulation (FEniCSx):
   • Import XDMF mesh.
   • Define function spaces.
   • Solve Stokes equations (Fluid flow).
   • Solve Advection-Diffusion (Transport).

---

4. Demonstration Data

• Gonzo: (GitHub: jorgenriseth/gonzo)
• OpenNeuro ds004478: Raw T1w and fMRI BOLD data.
• MNI152: Standard template for registration testing.

Other data / software at Zenodo

• csftracermodeling/csftracermodeling: Submitted code
• The Gonzo Dataset: Human Brain MRI Data of CSF Tracer Evolution Over 72h For Data-Integrated Simulations
• Ratbrain mesh
• jorgenriseth/mri2fem-multicompartment: v2.1
• meg-simula/optimal-transport-fenics: v1.0.1
• Software for Mathematical modeling of the human brain - from magnetic resonance images to finite element simulation
• Software and data for "Advection versus diffusion in brain ventricular transport"
• Example meshes for 'Human brain solute transport quantified by glymphatic MRI-informed biophysics during sleep and sleep deprivation'
• Software and data for "In-silico molecular enrichment and clearance of the human intracranial space"
• Software and data for "Stretch and flow at the gliovascular interface: high-fidelity modelling of the mechanics of astrocyte endfeet"
• Human intracranial pulsatility during the cardiac cycle: a computational modelling framework
• Supplementary material (code) for Are brain displacements and pressures within the parenchyma induced by surface pressure differences? A computational modelling study by E. Piersanti, M. E. Rognes, V. Vinje
• Uncertainty quantification of parenchymal tracer distribution using random diffusion and convective velocity fields (data sets)
• Mesh files for idealized and image-based periarterial and perivenous spaces
• mechanisms-behind-pvs-flow-v1.0
• geometrically-reduced-PVS-flow-v1.0
• MRI2FEM data set

---

5. Summary of repositories and other resources

• MRI2FEM book: paper, code
• MRI2FEM2 book: To be announced
• MariusCausemann/intracranialPulsation - Paper repo (Legacy FEnICS + multiphenics)
• MariusCausemann/brain-PVS-SAS-transport - Paper repo (Legacy FEnICS)
• gMRI2FEM - Tool for working with GMRI data (based on FEniCS, SVMTK and many other tools)
• pantarei - A collection of utilities for fluid flow modelling using FEniCS (legacy)
• Gonzo - Full pipline with dataset (based on gMRI2FEM)
• fenics-in-the-wild - Meshing with FtetWild
• scifem - Scientific finite element toolbox (also some biomedical) (FEniCSx)
• mri2mesh - Tools for converting images to mesh (PyVista + FEniCSx)
• bzapf/braintransport - Paper repo (Legacy FEniCS + dolfin-adjoint)
• cdaversin/CerebroSpinalFlow - Full pipeline in FEniCSx