Teach your computer to see objects on bioimages with Ilastik
Teach your computer to see objects on bioimages with Ilastik
to be announced
General context
Many experimental biologists and imaging facility users need to perform sophisticated segmentation of microscopy images, such as distinguishing between different cell structures or quantifying objects in dense, noisy backgrounds. Classical methods (thresholding, watershed, and manual annotation) often struggle with the variability and complexity of bioimage datasets. Ilastik provides an accessible machine-learning-based solution, allowing pixel- and object-level classification and integrating seamlessly with Fiji for downstream image processing, quantitative analysis, and batch processing across multiple images
- Explain the principles of supervised machine learning for pixel and object classification in microscopy images.
- Create and annotate pixel and object classifiers in Ilastik for simple segmentation and classification tasks.
- Export Ilastik probability maps and segmentations for further analysis in Fiji.
- Set up and use the Ilastik Fiji plugin for batch processing and automating pipelines for multi-image datasets
- Apply pre-trained models from the BioImage Model Zoo and adapt them to their data
This course uses a strong hands-on approach with guided exercises in Ilastik and Fiji
No formal programming or machine learning expertise is required, but you should:
- Be familiar with visualizing and processing microscopy images using Fiji. If you have no experience, you can follow our training on Basic image processing using Fiji.
- Have some basic understanding of the biological structure of interest (e.g., nuclei, cells, tissues) to inform annotation and classifier setup.
- Ilastik (latest release, Windows/ Mac/ Linux)
- FIJI with the Ilastik FIJI plugin (for direct integration/ export/ import, especially to handle large datasets and automate pipelines).
This course is part of the learning path Bioimaging.
AI methods used
This course applies supervised machine learning methods using Random Forest for image segmentation and classification, using pixel- and object-level models trained on user-provided annotations. Participants will work with pre-trained and custom models to generate probability maps and segmentations, without requiring in-depth knowledge of model development
