Tutorial about scaling-up image analysis with Dask presented during SciPy 2025 Conference (July 7-13, 2025)
This workshop uses Dask, Zarr, TiffFile, and Cellpose libraries.
- Dask is a parallel computing library in Python used for larger-than-RAM computation.
- The Zarr library allows us to interact with Next Generation File Formats, such as the
.zarrformat, which is highly compatible with large-scale images stored either on local disk, or remote cloud storage. - TiffFile is a library for opening
.tifimage files, and even other proprietary file formats. - Cellpose is a deep learning segmentation method for biological structures segmentation.
The most convenient way to set up the requires packages for this tutorial is by creating a conda environment.
conda create -n scale-up python=3.11 -c conda-forge
Install the following packages in the newly created environment.
conda activate scale-up
conda install -c conda-forge "cellpose>=3.0,<4.0" "tifffile>=2018.10.18,<=2025.5.21" "zarr>=2.0,<3.0" dask=2025.5.1 distributed=2025.5.1 dask-image=2024.5.3 imagecodecs=2025.3.30 requests=2.32.4 aiohttp=3.12.13
We'll open some image files that are compressed using JPEG2000, so we need imagecodecs to have access to this compression algorithm.
Additionally, install Jupyterlab to follow the notebooks, and its Dask extension.
conda install -c conda-forge jupyterlab=4.4.3 dask-labextension=7.0.0
We'll work with big microscopy images: whole slide images. These are images obtained by scanning slides on which thin slices of tissue have been mounted. A number of examples of these kind of image can be found in the OpenSlide test data, for example in the Aperio SVS format. These are TIFF variants that can be read with the Tifffile Python library.
For this workshop, we recommend downloading a smaller crop of the CMU-1 image from our google drive, along with its Zarr version.
A alternative even smaller example image can be found in the OpenSlide test data, and its corresponding Zarr version can also be downloaded from our google drive.
Please pre-download either pair of images before the tutorial!