WMH Segmentation (MARS-WMH nnU-Net)

Overview

MARS-WMH is the state-of-the-art, clinically-validated deep-learning tool (nnU-Net architecture) for segmenting brain white matter hyperintensities of presumed vascular origin. It takes a FLAIR image (recommended 1 mm isotropic) and a co-registered or registrable T1w image (1 mm isotropic, no contrast) and outputs a precise WMH segmentation mask in NIfTI format (returned in the original input resolution).

This skill serves as the NeuroClaw interface-layer wrapper for the official MARS-WMH Docker container (ghcr.io/miac-research/wmh-nnunet:latest) and strictly follows the hierarchical design:

1. Check whether Docker (with NVIDIA Container Toolkit) is installed (docker --version + nvidia-smi via claw-shell).
2. If nvidia-smi fails → immediately print the exact NVIDIA Container Toolkit installation commands and instruct the user to run them manually before retry.
3. If paths not provided → interactively ask the user for FLAIR and T1w full paths and confirm they exist on disk.
4. If paths provided → verify file existence and readability.
5. Prepare clean working directory, copy inputs, generate exact Docker pull/tag + run commands.
6. Generate a numbered execution plan.
7. Present the plan, estimated runtime (~5–15 min on GPU), requirements and risks → wait for explicit user confirmation (“YES” / “execute” / “proceed”).
8. On confirmation → delegate all shell execution to claw-shell.
9. Report completion, exact output mask location, and next steps.

Key design principle (2026 update): All Docker execution is routed through claw-shell.

Quick Reference (Common Use Cases)

Task	Recommended approach
Standard WMH segmentation	Default Docker run (nnU-Net, GPU)
Already co-registered images	Add --skipRegistration flag
CPU-only fallback	Remove --gpus all (slow)

Installation Check & Setup

Installation of Docker is delegated to dependency-planner.

GPU check performed before every run:

• If nvidia-smi fails, prompt user to run:

curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | \
  sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
curl -fsSL https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | \
  sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#' | \
  sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list > /dev/null
sudo apt-get update
sudo apt-get install -y nvidia-container-toolkit
sudo nvidia-ctk runtime configure --runtime=docker
sudo systemctl restart docker

Prerequisites:

• dependency-planner
• claw-shell
• NVIDIA GPU + drivers
• ≥8 GB VRAM

NeuroClaw recommended wrapper script

# WMH Segmentation Shell Commands (execute via claw-shell)

# 0. GPU check
nvidia-smi >/dev/null 2>&1 || {
  echo "GPU not detected. Run the following commands to install NVIDIA Container Toolkit:"
  cat << 'EOF'
curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | \
  sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
curl -fsSL https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | \
  sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#' | \
  sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list > /dev/null
sudo apt-get update
sudo apt-get install -y nvidia-container-toolkit
sudo nvidia-ctk runtime configure --runtime=docker
sudo systemctl restart docker
EOF
  exit 1
}

# 1. Pull & tag
docker pull ghcr.io/miac-research/wmh-nnunet:latest
docker tag ghcr.io/miac-research/wmh-nnunet:latest mars-wmh-nnunet:latest

# 2. Prepare workspace (replace paths with user-provided FLAIR/T1w)
FLAIR="/path/to/FLAIR.nii.gz"
T1="/path/to/T1w.nii.gz"
OUTPUT_DIR="wmh_output"
mkdir -p "$OUTPUT_DIR"
cp "$FLAIR" "$OUTPUT_DIR/FLAIR.nii.gz"
cp "$T1" "$OUTPUT_DIR/T1w.nii.gz"

# 3. Fix Docker data directory permission issues (common on Ubuntu)
chmod -R 777 "$OUTPUT_DIR"

# 4. Run inference
docker run --rm --gpus all \
  -v "$(pwd)/$OUTPUT_DIR:/data" \
  mars-wmh-nnunet:latest \
  --flair /data/FLAIR.nii.gz \
  --t1 /data/T1w.nii.gz

# 5. Show output mask
ls -lh "$OUTPUT_DIR"/*.nii*
echo "WMH segmentation mask saved in $OUTPUT_DIR"

Important Notes & Limitations

• Docker data directory permission issues are automatically fixed with chmod -R 777 on the output directory.
• If docker run fails with permission denied → run newgrp docker first, then retry.
• First run pulls image (~several GB); subsequent runs are fast.
• Input must be NIfTI; use dcm2nii if starting from DICOM.
• Output mask appears in $OUTPUT_DIR (exact filename shown by final ls).

When to Call This Skill

• User provides FLAIR + T1w and wants WMH segmentation.
• Any mention of MARS-WMH, nnU-Net WMH, white matter lesions segmentation.

Complementary / Related Skills

• dcm2nii → convert DICOM to NIfTI input
• dependency-planner → install Docker and NVIDIA Container Toolkit
• claw-shell → safe Docker execution

Reference

Official repo: https://github.com/miac-research/MARS-WMH Docker image: ghcr.io/miac-research/wmh-nnunet:latest Custom NeuroClaw skill.

Created At: 2026-03-23 Last Updated At: 2026-03-26 00:29 HKT Author: chengwang96