Democratising deep learning for microscopy with ZeroCostDL4Mic

Lucas von Chamier (1) , Romain F Laine (1) , Johanna Jukkala (2, 3) , Christoph Spahn (4) , Daniel Krentzel (5) , Elias Nehme (6) , Martina Lerche (2, 3) , Sara Hernández-Pérez (2, 3) , Pieta K Mattila (2, 3) , Eleni Karinou (7) , Séamus Holden (7) , Ahmet Can Solak (8) , Alexander Krull (9, 10) , Tim-Oliver Buchholz (9, 11) , Martin L Jones (12) , Loïc A Royer (8) , Christophe Leterrier (13) , Yoav Shechtman (6) , Florian Jug (11) , Mike Heilemann (4) , Guillaume Jacquemet (2, 3) , Ricardo Henriques (1)

1 UCL - University College of London [London]
2 University of Turku
3 Åbo Akademi University [Turku]
4 Goethe-University Frankfurt am Main
5 Imperial College London
6 Technion - Israel Institute of Technology [Haifa]
7 Newcastle University [Newcastle]
8 Chan Zuckerberg BioHub [San Francisco, CA]
9 Max Planck Institute for Molecular Biomedicine
10 Max Planck Institute for Physics
11 CSBD - Center for Systems Biology Dresden
12 The Francis Crick Institute [London]
13 INP - Institut de neurophysiopathologie

Deep Learning (DL) methods are powerful analytical tools for microscopy and can outperform conventional image processing pipelines. Despite the enthusiasm and innovations fuelled by DL technology, the need to access powerful and compatible resources to train DL networks leads to an accessibility barrier that novice users often find difficult to overcome. Here, we present ZeroCostDL4Mic, an entry-level platform simplifying DL access by leveraging the free, cloud-based computational resources of Google Colab. ZeroCostDL4Mic allows researchers with no coding expertise to train and apply key DL networks to perform tasks including segmentation (using U-Net and StarDist), object detection (using YOLOv2), denoising (using CARE and Noise2Void), super-resolution microscopy (using Deep-STORM), and image-to-image translation (using Label-free prediction-fnet, pix2pix and CycleGAN). Importantly, we provide suitable quantitative tools for each network to evaluate model performance, allowing model optimisation. We demonstrate the application of the platform to study multiple biological processes.

Mots clés: Cellular imaging; Image processing; Machine learning

Domaines: Neurobiologie

Lucas von Chamier, Romain F Laine, Johanna Jukkala, Christoph Spahn, Daniel Krentzel, et al.. Democratising deep learning for microscopy with ZeroCostDL4Mic. Nature Communications, 2021, 12, ⟨10.1038/s41467-021-22518-0⟩. ⟨hal-03233449⟩