May 10 2021
Researchers have begun to apply a method, known as light-field microscopy, to visualize the rapid neuronal signals in a fish brain. This method makes it viable to image these rapid biological processes in 3D.
A representation of a neural network provides a backdrop to a fish larva’s beating heart. Image Credit: Tobias Wüstefeld.
However, the images usually lack quality, and it takes hours or even days to convert enormous amounts of data into 3D volumes and movies.
Researchers from the European Molecular Biology Laboratory (EMBL) have now integrated artificial intelligence (AI) algorithms with two state-of-the-art microscopy methods—an advancement that reduces the time required to process the images from days to just seconds, while making sure that the resulting images are precise and sharp. The study results have been published in the Nature Methods journal.
Ultimately, we were able to take ‘the best of both worlds’ in this approach. AI-enabled us to combine different microscopy techniques, so that we could image as fast as light-field microscopy allows and get close to the image resolution of light-sheet microscopy.
Nils Wagner, Study Co-Lead Author and PhD Student, Technical University of Munich
While light-field microscopy and light-sheet microscopy sound almost the same, these methods come with different benefits and difficulties. Light-field microscopy captures huge 3D images that enable scientists to monitor and quantify incredibly fine movements, like the beating heart of a fish larva, at extremely high speeds. However, this method generates enormous amounts of data, which can take several days to process, and the ultimate images typically lack resolution.
Light-sheet microscopy tracks down a single 2D plane of a specified specimen at a time, and hence scientists could image specimens at higher resolution. In contrast to light-field microscopy, light-sheet microscopy generates images that can be processed faster; however, the data is not quite extensive because they simply capture data from a solo 2D plane at a time.
To leverage the advantages of each method, the EMBL team developed a method that employs light-sheet microscopy to train the AI algorithms and light-field microscopy to image huge 3D samples, which subsequently produce a precise 3D picture of the spacemen.
If you build algorithms that produce an image, you need to check that these algorithms are constructing the right image.
Anna Kreshuk, Group Leader, European Molecular Biology Laboratory
Kreshuk’s group brought machine learning expertise to the study. In the latest study, the team used light-sheet microscopy to ensure that the AI algorithms were functioning.
“This makes our research stand out from what has been done in the past,” Kreshuk added.
Robert Prevedel, the group leader from EMBL and whose team contributed to the new hybrid microscopy platform, observed that the real bottleneck in constructing improved microscopes is not optics technology as assumed generally, but rather computation. That is the reason why, back in 2018, Prevedel and Anna decided to team up.
Our method will be really key for people who want to study how brains compute. Our method can image an entire brain of a fish larva, in real time.
Robert Prevedel, Group Leader, European Molecular Biology Laboratory
According to Prevedel and Anna, this method could also be altered to work with other types of microscopes, ultimately enabling biologists to look at scores of different samples and observe much more and relatively quicker. For instance, it could help identify genes that play a role in heart development, or could quantify the activity of scores of neurons simultaneously.
The team has now planned to investigate whether the technique can be used on bigger species, such as mammals.
Fynn Beuttenmüller, the co-lead author of the study and a PhD student in the Kreshuk group at EMBL Heidelberg, clearly believes in the power of AI. “Computational methods will continue to bring exciting advances to microscopy,” concluded Beuttenmüller.
Journal Reference:
Wagner, N., et al. (2021) Deep learning-enhanced light-field imaging with continuous validation. Nature Methods. doi.org/10.1038/s41592-021-01136-0.