shreyag_ibd_orgs

Analysis of colon organoids in 3D: Cellpose-based segmentation, organoid detection, and per-cell/per-organoid measurements from Nikon .nd2 images. Inputs (markers, channels, compartments) are set in config.yaml or via CLI.

Requirements: pixi, Git and a CUDA-capable GPU (for Cellpose and OpenCL).

---

Running locally

From the project root, install the environment and run the pipeline.

If you are not familiar with command line the easiest will be to use Jupyter Notebooks, just copy and paste the following command in your terminal (i.e. cmd for Windows):

git clone https://github.com/adiezsanchez/shreyag_ibd_orgs.git && cd shreya_ibd_orgs && pixi run lab

This will clone the repo, install the virtual environment via pixi and launch a Jupyter instance to interact with the Jupyter Notebooks inside ./notebooks.

Process a single image

pixi run python main.py --image "/path/to/image.nd2" --config config.yaml

Or after entering the environment:

pixi shell
python main.py --image "/path/to/image.nd2" --config config.yaml

Process folder containing images (all .nd2 in a directory)

Use pixi tasks so the environment is used automatically.

Cross-platform (pixi picks the script by OS):

pixi run run_folder "/path/to/image/folder"
pixi run run_folder "/path/to/image/folder" other_config.yaml

Arguments: first = image folder path (required, remember quotation marks " " ), second = config file (optional, default config.yaml).

---

Jupyter notebooks

Interactive batch runs and exploration:

pixi run lab

Then open the notebooks under notebooks/ (e.g. BP_organoid_analysis.ipynb, SP_organoid_analysis.ipynb). They loop over images in a folder and run the same pipeline logic interactively.

---

IDUN HPC (NTNU)

On IDUN, use Slurm to process a directory of images in parallel on GPUs. From the project directory on the cluster:

# Process all .nd2 in a directory (default config: config.yaml)
sbatch hpc_sbatch/run_pipeline_slurm.sh /path/to/directory_with_nd2

# Optional: specify config path
sbatch hpc_sbatch/run_pipeline_slurm.sh /path/to/directory_with_nd2 config.yaml

Example with project data path:

sbatch hpc_sbatch/run_pipeline_slurm.sh /cluster/work/albertds/raw_data/20260114_T7_2microns

Check job status:

squeue -u $USER

Edit hpc_sbatch/run_pipeline_slurm.sh to set PROJECT_DIR and Slurm account/partition if needed. Other scripts in hpc_sbatch/ target different GPU types (A100, H100, P100, V100) just for testing OpenCL and CUDA acceleration.