Deep-ROCS: from speckle patterns to superior-resolved images by deep learning in rotating coherent scattering...

Area of Science:

• Optics and Photonics
• Biomedical Imaging
• Computational Imaging

Background:

• Rotating coherent scattering (ROCS) microscopy offers label-free imaging beyond the optical diffraction limit.
• Conventional ROCS achieves 150 nm spatial and 10 ms temporal resolution but may lose sample information due to speckle pattern summation.

Purpose of the Study:

• To introduce Deep-ROCS, a neural network-based technique for superior-resolution image reconstruction.
• To improve the accuracy and information retrieval of ROCS microscopy.

Main Methods:

• Development of a neural network (Deep-ROCS) for efficient numerical combination of ROCS images.
• Utilizing a small set of differently illuminated speckle images (n=6) for reconstruction.

Main Results:

• Deep-ROCS reconstructs super-resolved images with higher accuracy than conventional ROCS.
• The technique successfully retrieves high-frequency information from limited speckle images.
• Experimental validation on 200 nm beads and simulations on filament networks demonstrated performance.

Conclusions:

• Deep-ROCS significantly advances ROCS microscopy by enabling more accurate super-resolution imaging.
• The method shows potential for imaging complex biological structures with enhanced detail.