Bacterial bioimage analysis using open-source deep learning approaches
- Abstract number
- 39
- Presentation Form
- Oral
- DOI
- 10.22443/rms.elmi2021.39
- Corresponding Email
- [email protected]
- Session
- Image Data Analysis, Management and Visualisation
- Authors
- Christoph Spahn (2, 8), Pedro Matos Pereira (3), Romain Laine (4), Mia Conduit (1), Lucas von Chamier (4), Mariana Pinho (3), Séamus Holden (1), Guillaume Jacquemet (5, 6), Mike Heilemann (2), Ricardo Henriques (4, 7)
- Affiliations
-
1. Centre for Bacterial Cell Biology, Newcastle University Biosciences Institute, Faculty of Medical Sciences, Newcastle upon Tyne NE24AX, United Kingdom
2. Institute of Physical and Theoretical Chemistry, Goethe-University Frankfurt, 60439 Frankfurt, Germany
3. Instituto de Tecnologia Química e Biológica António Xavier, Universidade Nova de Lisboa, Oeiras, Portugal
4. MRC-Laboratory for Molecular Cell Biology, University College London, London WC1E 6BT, United Kingdom
5. Turku Bioscience Centre, University of Turku and Åbo Akademi University, Turku, Finland
6. Faculty of Science and Engineering, Cell Biology, Åbo Akademi University, Turku, Finland
7. Instituto Gulbenkian de Ciência, 2780-156 Oeiras, Portugal
8. Max-Planck-Institute for Terrestrial Microbiology Department Natural Products in Organismic Interactions, 35043 Marburg, Germany
- Keywords
Microbiology, Deep Learning, Super-resolution microscopy, Antibiotic research
- Abstract text
Deep Learning (DL) is rapidly changing the field of microscopy and enables complex data analysis outperforming classical algorithms [1]. This revolution is accompanied by a large effort in creating open-source tools that also allow non-computer scientists to effectively use this powerful technology [2]. The main use of such tools is the analysis of microscopy data obtained from eukaryotic samples, but they are hardly used for microbiological research despite their great potential. In this work, we exploit the capacity of state-of-the-art artificial neural-networks to data-mine bacterial microscopy images. For this purpose, we created a database of training and testing data, that (i) helps bacteriologists new to the machine-learning field to rapidly learn and start exploiting how to analyze their data and (ii) introduces innovative applications of DL for different areas in microbiology such as live- and single-cell bacteriology or antibiotic research. To do so, we matched computational methods we recently developed in the ZeroCostDL4Mic platform [2] to this data, establishing a highly accessible image analysis platform that can rapidly be used by researchers studying microbiology. Using this database, we demonstrate segmentation of brightfield and fluorescence images of different bacterial species [3], identify cell division events in time-lapse imaging and classify phenotypes of antibiotic-treated cells [4]. We further use networks to denoise live-cell microscopy data, artificially label brightfield images for cell membranes and predict SIM super-resolution images from widefield fluorescence signal [5]. These examples demonstrate the breadth of possible applications for deep learning technology in microscopy-based microbiology and can be used as a starting point for bacteriologists to train their own models.
- References
[1] LeCun, Y., Bengio, Y. & Hinton, G. Deep learning. Nature 521, 436–444 (2015)
[2] von Chamier, L. et al. ZeroCostDL4Mic: an open platform to use Deep-Learning in Microscopy bioRxiv 2020.03.20.000133 (2020)
[3] Schmidt, U., Weigert, M., Broaddus, C. & Myers, G. Cell Detection with Star-convex Polygons. arXiv 1806.03535 (2018)
[4] Redmon, J. & Farhadi, A. YOLO9000: Better, Faster, Stronger. arXiv 1612.08242 (2016)
[5] Weigert, M., Schmidt, U., Boothe, T. et al. Content-aware image restoration: pushing the limits of fluorescence microscopy. Nat Methods 15, 1090–1097 (2018)