AI technology produces clear images of thick biological samples without additional hardware.

Depth degradation is a problem that biologists are familiar with. The deeper you examine the sample, the blurrier the image becomes. C. elegans embryos and pieces of tissue are only tens of microns thick, but when the instrument peeks past the top layer, light bending causes the microscopic image to lose its sharpness.

To address this problem, microscopists add techniques to existing microscopes that counteract these distortions. But this technique, called adaptive optics, requires time, money and expertise, making it available to relatively few biological labs.

Now researchers at HHMI’s Janelia Research Campus and their collaborators have made a similar correction without using adaptive optics, adding additional hardware, or taking more images. developed a method. A team at the Shroff Institute has developed a new AI method that produces clear microscopic images across thick biological samples.

The paper will be published in the journal Nature Communications.

To create the new technique, the team first figured out how to model how the image degrades when a microscope images deep inside a homogeneous sample. When we then applied the model to undegraded foreground images of the same sample, these sharp images were distorted in the same way as the back images. They then trained a neural network to reverse the distortion across the sample, resulting in sharper images throughout the depth of the sample.

In addition to producing better-looking images, this method also allowed the researchers to more accurately count the number of cells in C. elegans embryos, trace blood vessels and ducts throughout mouse embryos, and trace parts of mouse livers and hearts. It is now also possible to examine the mitochondria in the body.

New deep learning-based techniques are more accessible than traditional adaptive optics techniques because they require no equipment other than a standard microscope, a computer with a graphics card, and a short tutorial on how to run computer code. I am.

The Shroff lab is already using the new technique to image C. elegans embryos, and the team can further develop the model to reduce its dependence on sample structure and apply the new method to less homogeneous samples. I plan to do so.