DL4MicEverywhere: Deep learning for microscopy made flexible, shareable, and reproducible

Session

Poster Session

Authors

Iván Hidalgo-Cenalmor (4), Joanna W Pylvänäinen (3), Mariana G Ferreira (4), Craig T Russell (2), Ignacio Arganda-Carreras (1, 5, 8), Guillaume Jacquemet (3, 6, 9, 10), Ricardo Henriques (4, 7), Estibaliz Gómez-de-Mariscal (4)

Affiliations

1. Dept. Computer Science and Artificial Intelligence, University of the Basque Country (UPV/EHU)
2. European Molecular Biology Laboratory, European Bioinformatics Institute, Wellcome Genome Campus
3. Faculty of Science and Engineering, Cell Biology, Åbo Akademi University, Turku
4. Optical cell biology group, Instituto Gulbenkian de Ciência, Oeiras
5. IKERBASQUE, Basque Foundation for Science
6. Turku Bioscience Centre, University of Turku and Åbo Akademi University, Turku
7. UCL Laboratory for Molecular Cell Biology, University College London, London
8. Donostia International Physics Center (DIPC)
9. Turku Bioimaging, University of Turku and Åbo Akademi University, Turku
10. InFLAMES Research Flagship Center, Åbo Akademi University, Turku

Keywords

deep learning

bioimage analysis

accessible

reproducible

no-code

Abstract text

Deep learning has revolutionised how researchers process microscopy images, providing scalable and highly accurate automatic methods to analyse extensive microscopy data. Yet adopting these advanced techniques is challenging for biologists lacking expertise in this field1,2. Various user-friendly tools, including deepImageJ3, Ilastik4, the BioImage Model Zoo5, and our ZeroCostDL4Mic6 platform, have made deep learning more approachable. ZeroCostDL4Mic contributed with a collection of user-friendly non-code Jupyter Notebooks that share the same structure for data loading, training, testing and deployment of deep learning models and easily connect with Google Colab’s free GPU services. This way, it enabled user-friendly deep learning experience for several image processing tasks such as segmentation, object detection, denoising, super-resolution microscopy, and image-to-image translation. However, these deep learning pipelines rely on intricate and very specific Python environment installations, impeding their long-term reproducibility.

To solve this issue, we propose DL4MicEverywhere (https://github.com/HenriquesLab/DL4MicEverywhere). DL4MicEverywhere is a community-driven platform that allows the encapsulation of user-friendly deep learning image processing pipelines within Docker7 images without dealing with the installation process. These images serve as standalone virtualisations of required packages and code to reproduce a computational environment, making them transferable across systems such as local computers, high performance clusters, or cloud services. 

DL4MicEverywhere builds upon ZeroCostDL4Mic to create easy-to-use interfaces based on codeless Jupyter Notebooks. Additionally, a graphical user interface isolates the end-user from the containerisation and environment set-up process when launching their pipelines, permitting others to replicate analysis, benchmark methods and build on research across different systems (e.g., local computers, cloud computing). Namely, DL4MicEverywhere combines an automatic containerisation process with version tracking, achieved through unique Docker tags, to enhance the stability8 of image processing tasks. For this, we propose a new containerisation standard, compatible with the format of the BioImage Model Zoo and connected with an automatic continuous integration pipeline9, that facilitates developers’ contributions to our new platform with minimal effort.

 

In brief, DL4MicEverywhere is a flexible platform that facilitates the contribution, reproducibility and dissemination of deep learning approaches. It empowers non-expert end-users with an easy-to-use tool to replicate deep learning-based image processing pipelines with their data anywhere, without the intricacies of dependencies’ installation. We expect DL4MicEverywhere to become a baseline for a more reproducible and stable AI in bioimaging.

References

1. Moen, E. et al. Deep learning for cellular image analysis. Nature Methods 16, 1233–1246 (2019). 

2. Pylvänäinen, J. W., Gómez-de-Mariscal, E., Henriques, R. & Jacquemet, G. Live-cell imaging in the deep learning era. Current Opinion in Cell Biology 85, 102271 (2023). 

3. Gómez-de-Mariscal, E. et al. DeepImageJ: A user-friendly environment to run deep learning models in ImageJ. Nat Methods 18, 1192–1195 (2021). 

4. Berg, S., Kutra, D., Kroeger, T. et al. ilastik: interactive machine learning for (bio)image analysis. Nat Methods 16, 1226–1232 (2019). https://doi.org/10.1038/s41592-019-0582-9

5. Ouyang, W. et al. BioImage Model Zoo: A Community-Driven Resource for Accessible Deep Learning in BioImage Analysis. bioRxiv 2022.06.07.495102 (2022) doi:10.1101/2022.06.07.495102. 

6. von Chamier, L. et al. Democratising deep learning for microscopy with ZeroCostDL4Mic. Nature Communications 12, 2276 (2021). 

7. Merkel, D. Docker: lightweight linux containers for consistent development and deployment. Linux j 239, 2 (2014). 

8. Moreau, D., Wiebels, K. & Boettiger, C. Containers for computational reproducibility. Nat Rev Methods Primers 3, 1–16 (2023). 

9. Beaulieu-Jones, B. K. & Greene, C. S. Reproducibility of computational workflows is automated using continuous analysis. Nat Biotechnol 35, 342–346 (2017).