Tutorial 1 — Setting Up Your Environment

AI Imaging Workshop · WSUSOM · the 9:30 AM "Python / SLURM / Good Programming Practices" block.

By the end of this tutorial you will have, on the WSU grid (and/or your own laptop):

1. Miniforge — Python plus mamba, a fast package manager that keeps each project's software isolated and reproducible.
2. An imaging environment with the libraries we use to read and manipulate CT scans and segmentation labels (nibabel, SimpleITK, pydicom, …).
3. Nextflow — the tool we use to run reproducible, parallel pipelines on the SLURM cluster.

You only do this once. Everything is copy-paste; read the one-line explanation above each block so you know what you are running, not just that it runs.

No grid account? Use your laptop — it all works the same. Every step below runs identically on your own machine; you just install Miniforge locally instead of on the grid. The only thing that changes per-platform is the installer file in Step 1 (Linux / macOS / Windows-WSL), which we call out there. If you do have a grid account, install on the grid so your environment sits right next to the GPUs you'll use in Tutorial 5.

0. Open a terminal

Everything below is typed into a terminal (a text window where you run commands).

On the WSU grid (replace youraccessid with your WSU AccessID):

ssh youraccessid@grid.wayne.edu

You know you are in the right place when each line starts with a prompt like [youraccessid@warrior ~]$.

On your own laptop (no grid account needed):

• macOS — open the Terminal app (Applications → Utilities → Terminal).
• Linux — open your terminal app.
• Windows — install WSL (Windows Subsystem for Linux) once: open PowerShell and run wsl --install, reboot if asked, then open the Ubuntu app. That gives you a Linux terminal where every command below works unchanged.

1. Install Miniforge (Python + mamba)

What it is: Miniforge bundles Python and mamba — a fast, drop-in replacement for conda that downloads software libraries and keeps them in self-contained "environments" so two projects can never break each other's dependencies. It defaults to the open conda-forge channel, so there's nothing extra to configure. (mamba and conda are interchangeable and share the same environments; we use mamba because it resolves dependencies much faster.)

Download the installer (Linux x86-64 — this is the grid and most laptops):

cd ~
curl -L -o miniforge.sh \
  https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh

On a Mac, swap the installer for your chip:

• Apple Silicon (M1/M2/M3/M4): Miniforge3-MacOSX-arm64.sh
• Intel Mac: Miniforge3-MacOSX-x86_64.sh

(Same URL prefix, just change the filename at the end. On Windows, you're inside WSL/Ubuntu, so use the Linux installer above.)

Run the installer (-b = accept defaults, -p = where to put it):

bash miniforge.sh -b -p ~/miniforge3
rm miniforge.sh                     # tidy up the installer

Turn it on so your shell knows where to find mamba, now and in every future session:

~/miniforge3/bin/conda init bash
source ~/.bashrc                    # reload the shell so 'mamba' works now

You should now see (base) at the start of your prompt — that is the default environment, telling you it is active.

2. Create the imaging environment

Why a separate environment: we keep the workshop's libraries in their own named space (imaging) so that installing or upgrading something later never disturbs anything else. Activating it is how you "enter" that space.

Create it with Python 3.11:

mamba create -n imaging python=3.11 -y
mamba activate imaging

Your prompt now shows (imaging) instead of (base) — you are inside the new environment. (Run mamba activate imaging at the start of any future session.)

Install the imaging libraries — these are the tools for reading and editing CT scans and label maps:

mamba install -y \
  numpy scipy pandas matplotlib \
  nibabel pydicom simpleitk scikit-image

What each one is for:

3. Install Nextflow (for running pipelines on SLURM)

What it is: Nextflow runs multi-step analysis pipelines and submits the work to the SLURM scheduler for you — in parallel, and reproducibly (the same command gives the same result, on your laptop or the cluster). Nextflow needs Java, which mamba installs alongside it:

mamba install -y -c bioconda nextflow

That single command pulls in both Nextflow and a compatible Java runtime into your imaging environment.

4. Check that everything works

Python libraries — this prints the version of each and a final "OK":

python - <<'PY'
import numpy, scipy, pandas, matplotlib, nibabel, pydicom, SimpleITK, skimage
print("numpy     ", numpy.__version__)
print("scipy     ", scipy.__version__)
print("nibabel   ", nibabel.__version__)
print("pydicom   ", pydicom.__version__)
print("SimpleITK ", SimpleITK.__version__)
print("skimage   ", skimage.__version__)
print("ALL IMPORTS OK")
PY

Nextflow and Java:

nextflow -version
java -version

If you see version numbers and ALL IMPORTS OK, your environment is ready.

A 30-second real test — make a tiny 3-D volume, save it as NIfTI, read it back. This is exactly the read/write loop you'll use on real scans:

python - <<'PY'
import numpy as np, nibabel as nib
vol = np.random.randint(0, 100, size=(8, 8, 8)).astype(np.int16)   # fake volume
nib.save(nib.Nifti1Image(vol, affine=np.eye(4)), "test.nii.gz")    # write
back = np.asarray(nib.load("test.nii.gz").dataobj)                 # read
print("round-trip identical:", np.array_equal(vol, back))
PY
rm -f test.nii.gz

round-trip identical: True means you can load, edit, and save medical images.

5. Everyday use (after today)

Each time you log in and want to work, just activate the environment:

mamba activate imaging

To leave it:

mamba deactivate

To see your environments, or remove one and start over:

mamba env list                      # list all environments
mamba env remove -n imaging         # delete 'imaging' if you want a clean redo

Troubleshooting

• mamba: command not found after Step 1 → you missed source ~/.bashrc, or open a fresh terminal. As a fallback: source ~/miniforge3/bin/activate.
• mamba activate says "run mamba init first" → run ~/miniforge3/bin/conda init bash then source ~/.bashrc. (conda init sets up the shell hook that both mamba and conda use.)
• A package can't be found → it may live on another channel; add it for that one install, e.g. mamba install -y -c bioconda <pkg> (as we do for Nextflow).
• nextflow: command not found → make sure imaging is active (mamba activate imaging); Nextflow was installed into that environment.
• Out of disk space in your home directory on the grid → environments are large. Ask whether to place ~/miniforge3 on a project/scratch filesystem instead of home. (On a laptop this is rarely an issue.)

Next tutorial: DICOM vs. NIfTI — what's actually inside a medical image file, where we open a real CT scan and its segmentation with the tools you just installed.