Biophysicists Bring Us Closer to Intelligent Microscopes

Whenever someone wants to obtain detailed observations of bacterial division from a sample of live bacteria, things can get a bit complicated. They might have to stay at the microscope non-stop until the bacterium divides, which can take hours. Manual detection and acquisition control is actually very common in the field.

Another option is to set the microscope to take images indiscriminately and as often as possible, but excessive light can cause problems. It depletes the fluorescence from the sample quicker, which can prematurely destroy living samples. At the same time, there would be many unnecessary images generated, and only a few would actually contain images of dividing bacteria.

Yet one more solution is to use artificial intelligence (AI) to detect precursors to bacterial division and use them to automatically update the microscope’s control software, which would help it take more pictures of the division.

Automating Microscope Control

Looking at these three different options, a team of biophysicists at Ecole Polytechnique Fédérale de Lausanne (EPFL) have come up with a way to automate microscope control for imaging biological events in detail. And at the same time, the method limits stress on the sample. The new technique relies on artificial neural networksand it works for both bacterial cell division and mitochondrial division.

The team published their findings in Nature Methods.

Suliana Manley is principal investigator from EPFL’s Laboratory of Experimental Biophysics.

“An intelligent microscope is kind of like a self-driving car. It needs to process certain types of information, subtle patterns that it then responds to by changing its behavior, ”says Manley. “By using a neural network, we can detect much more subtle events and use them to drive changes in acquisition speed.”

The team first found a solution to detecting mitochondrial division, which is more difficult than a solution for certain bacteria. Mitochondrial division occurs more infrequently, meaning it is unpredictable, and it can happen almost anywhere within the mitochondrial network at any moment.

Training the Neural Network

The team trained the neural network to look for mitochondrial constrictions, which is a change in the shape of mitochondria that leads to division. They also observed a protein known to be enriched at sites of division.

The microscope will switch into high-speed imaging when both constrictions and protein levels are high, which enables it to capture many images of division events. But when the levels are low, the microscope will switch to low-speed imaging, which helps avoid exposing the sample to excessive light.

An intelligent fluorescent microscope such as this enables scientists to observe samples for longer compared to standard fast imaging. The sample was more stressed compared to standard slow imaging, but the team could obtain more meaningful data.

“The potential of intelligent microscopy includes measuring what standard acquisitions would miss,” Manley explains. “We capture more events, measure smaller constrictions, and can follow each division in greater detail.”

The team is now making the control framework available as an open-source plug-in for the open microscope software Micro-Manager. They wish to enable other scientists to integrate AI into their own microscopes.