Multicolor localization microscopy and point-spread-function engineering by deep learning.

Area of Science:

• Microscopy and Imaging
• Computational Biology
• Machine Learning Applications

Background:

• Deep learning excels in image classification and restoration.
• Microscopy often requires specialized optical elements for spectral classification.
• Existing methods for color classification in microscopy can be complex and costly.

Purpose of the Study:

• To apply deep learning to microscopy for color classification of single emitters.
• To demonstrate a neural network's ability to exploit chromatic aberration for spectral analysis.
• To develop a more efficient and less complex method for spectral classification in microscopy.

Main Methods:

• Utilized deep learning, specifically neural networks, for image analysis.
• Exploited the chromatic dependence of the point-spread function (PSF) in imaging.
• Employed a standard, unmodified single-channel grayscale camera configuration.

Main Results:

• Neural networks successfully classified colors of single emitters with high efficiency.
• The method accurately identified both static and mobile emitters.
• Deep learning facilitated the design of new phase-modulating elements for improved color differentiation, enabling simultaneous differentiation of four species.

Conclusions:

• Deep learning offers a powerful, hardware-independent approach to spectral classification in microscopy.
• This method simplifies multicolor imaging by removing the need for additional optical components.
• The developed neural network-based strategy significantly advances multicolor imaging capabilities in microscopy.