Summary:
UCLA researchers in the Department of Electrical Engineering have developed a novel high-speed and non-iterative approach of achieving super-resolution in fluorescence microscopy using deep learning.
Background:
Super-resolution microscopy such as structured illumination microscopy allows visualization of intricate details of cellular features and processes. These imaging modalities often require complex optics, specific fluorophores, labor-intensive experimental setup and extensive computational processing. Typical approaches of achieving super-resolution utilize a deterministic model restrictive to a specific imaging setup, and heavily rely on the accuracy of the underlying numerical image formation model. Therefore, these methods are not widely adaptive across imaging modalities.
Innovation:
A novel deep-leaning based method was created to achieve super-resolution in fluorescence microscopy. Unlike conventional super-resolution techniques, this method does not require input of any image processing mathematical models or excess optical parameters. It is solely based on training a generative adversarial network. This method is high-speed, non-iterative and applicable to images/sample types that the network is not trained for. In demonstrations, wide-field blurry images and diffraction-limited confocal images were all transformed by this approach into super-resolutions, matching images taken by high numerical aperture objectives and stimulated emission depletion (STED) microscopes.
Potential Applications:
• Simulated emission depletion microscopy
• Wide-field fluorescence microscopy
• Confocal fluorescence microscopy
• Structured illumination microscopy
• Total internal reflection fluorescence microscopy
• Other fluorescence imaging techniques
Advantages:
• Widely adaptive across imaging modalities
• High-speed
• Non-iterative
• Requires no prior knowledge about sample preparations and image formation models
Patent:
Systems and methods for deep learning microscopy
Related Papers:
Wang, H., Rivenson, Y., Jin, Y. et al. Deep learning enables cross-modality super-resolution in fluorescence microscopy. Nat Methods 16, 103–110 (2019). https://doi.org/10.1038/s41592-018-0239-0
Reference:
UCLA Case No. 2018-739
Lead Inventor:
Aydogan Ozcan