Deep learning framework for rapid aberration correction in reflection matrix microscopy.

Area of Science:

• Optical Imaging
• Biomedical Optics
• Machine Learning in Optics

Background:

• Scattering media in optical imaging cause angle-dependent phase distortions, degrading image quality.
• Aberrations in reflection matrix microscopy limit resolution and speed.
• Current aberration correction methods can be computationally intensive.

Purpose of the Study:

• To present a novel deep learning-based method for correcting aberrations in reflection matrix microscopy.
• To enhance image quality and enable real-time, high-resolution imaging.
• To improve the computational efficiency of aberration correction.

Main Methods:

• A U-Net-based deep learning model was trained to predict and correct input and output aberrations directly from the reflection matrix.
• An iterative inference and correction process was employed to eliminate round-trip aberrations.
• A covariance matrix-based training strategy was introduced to optimize efficiency and reduce iteration time.

Main Results:

• The method effectively corrects round-trip aberrations, significantly enhancing imaging quality.
• A 100-fold computational speedup was achieved compared to conventional wave correlation algorithms.
• The approach demonstrated robustness across various aberration conditions in simulations and experiments.

Conclusions:

• The deep learning framework enables real-time, label-free imaging by overcoming computational bottlenecks.
• This method paves the way for rapid, high-resolution imaging in biomedical applications.
• The iterative correction and efficient training strategy significantly advance optical imaging capabilities.