Dockerized Setup for DeepWMH Segmentation Tool

This guide provides instructions to build and run the DeepWMH brain segmentation tool inside a Docker container.

---

1. Clone Repository and Navigate to Docker Directory

Clone the main repository locally and move into the Docker setup folder:

cd docker

---

2. Build the Docker Image

Optionally configure the Dockerfile, then build the Docker image using:

docker build -t <image_name> .

Replace <image_name> with your preferred image tag (e.g., deepwmh:predict).

---

3. Run the Segmentation Tool

There are two ways to run the segmentation pipeline:

Option A: Using the run_deepwmh.sh Script

This script processes images and generates logs automatically.

./run_deepwmh.sh

• Logs will be saved to the logs/ directory.
• Output will be stored in the output/ directory.

Option B: Run Docker Manually

You can also run the tool directly from the command line:

docker run --rm --gpus all \
    -v /path/to/data:/data \
    -v /path/to/output:/output \
    deepwmh:predict \
    -i /data/<flair_image> \
    -n <subject_id> \
    -m /model \
    -o /output/<subject_id> \
    -g 0 &

Make sure to replace:

• /path/to/data with the path to your FLAIR input images.
• /path/to/output with the desired output directory.
• <flair_image>, <subject_id>, and /model with your specific filenames/paths.

---

4. Debugging Mode

If you need to troubleshoot or access the container interactively:

Build a Debug Image with Bash Entrypoint

docker build -t deepwmh:shell --build-arg ENTRYPOINT="/bin/bash" .

Start the Container

docker run --gpus all --restart always -dit --name deepwmh deepwmh:shell

Access the Container Shell

docker exec -it deepwmh /bin/bash

---

Notes

• Make sure NVIDIA drivers and CUDA are correctly installed on your host machine.
• Use install_services.sh to set up and download required services.
• The model .zip file should be placed in the docker/ directory before building the image. It will be copied into the container during the build process.