Reviewed by Lexie CornerJan 8 2025
In a study published in Nature Portfolio, Janelia researchers developed an AI-based strategy to produce sharp microscopy images throughout thick biological samples.
Depth degradation is a common challenge for biologists: the deeper a sample is examined, the more the image quality deteriorates. Even in relatively thin samples, such as worm embryos or tissue sections that are only tens of microns thick, light bending distorts the image when viewing beyond the top layer.
Microscopists often address this issue using adaptive optics, a technique that corrects distortions by integrating specialized technology into microscopes. However, adaptive optics is costly, time-consuming, and requires expertise, limiting its use to a small number of advanced labs.
The researchers have now devised a method to achieve similar corrections without the need for additional hardware, adaptive optics, or extra image capturing.
To develop the technique, the team first simulated the image degradation that occurs when a microscope probes deeper into a uniform sample. They then used this model to distort clear, near-surface images of the same sample, mimicking the appearance of deeper layers. A neural network was trained on these distorted images to reverse the degradation, producing sharp images throughout the sample’s depth.
This technique not only generated clearer images but also enabled more accurate biological analyses. The researchers could precisely count cells in worm embryos, track veins and neural tracts in entire mouse embryos, and analyze mitochondria in mouse liver and heart fragments.
Unlike adaptive optics, this deep learning-based technology requires no additional equipment beyond a standard microscope, a computer with a graphics card, and basic training to run the software. This makes it far more accessible to most biology labs.
The Shroff Lab has already adopted the technology to study worm embryos. The team plans to further refine the model to reduce its dependence on uniform sample structures, enabling its application to more complex and varied samples.
New AI technique generates clear images of thick biological samples without the fancy hardware
Researchers have developed a new AI method that produces sharp microscopy images throughout thick biological samples. This video shows how the method, DeAbe, restores highly dynamic time-lapse images of live roundworms expressing a GCaMP6 marker targeted to neurons acquired with instant Structured Illumination Microscopy. The top video shows the raw images and the bottom video shows restoration after DeAbe, which allowed the researchers to resolve structural details in the nerve ring that were obscured in the raw data. Video Credit: Guo et al.
Journal Reference:
Guo, M., et. al. (2024) Deep learning-based aberration compensation improves contrast and resolution in fluorescence microscopy. Nature Portfolio. doi.org/10.1038/s41467-024-55267-x