Auto-focusing and quantitative phase imaging using deep learning for the incoherent illumination microscopy system.

Area of Science:

• Biomedical Optics
• Microscopy
• Computational Imaging

Background:

• Quantitative phase information is crucial for biomedical studies but difficult to obtain with traditional bright-field microscopy.
• Maintaining focus in living samples during long-term observation is a significant challenge in microscopy.

Purpose of the Study:

• To develop a simultaneous autofocusing and quantitative phase imaging technique for microscopy.
• To enable high-speed, label-free, and continuous monitoring of living cells.

Main Methods:

• A lightweight deep learning framework utilizing residual structures and a novel loss function was proposed.
• The framework processes single-shot out-of-focus bright-field images to reconstruct in-focus amplitude and phase information.
• Training data were acquired using a hybrid incoherent and coherent illumination system.

Main Results:

• The framework achieved high-speed reconstruction (10 frames per second) of focused amplitude and phase information.
• Experimental results demonstrated successful reconstruction of focused and quantitative phase images for both non-biological and biological samples.
• The method proved effective for continuous monitoring of living cells.

Conclusions:

• The proposed deep learning framework offers a versatile solution for simultaneous autofocusing and quantitative phase imaging.
• This technique enhances the capabilities of traditional incoherent illumination microscopy for label-free, long-term live-cell imaging.