Robust Phase Unwrapping via Deep Image Prior for Quantitative Phase Imaging.

Area of Science:

• Biomedical Optics
• Computational Imaging
• Cell Biology

Background:

• Quantitative Phase Imaging (QPI) is a label-free microscopy technique providing morphological and dynamic information.
• Phase wrapping is a common artifact in QPI, necessitating phase unwrapping for accurate image reconstruction.
• Unwrapping phase in thick, complex biological samples like organoids presents significant challenges.

Purpose of the Study:

• To develop a deep learning-based phase unwrapping method for QPI that eliminates the need for training datasets.
• To address the limitations of supervised deep learning approaches in handling complex biological samples.
• To enable reliable phase imaging of challenging samples using QPI.

Main Methods:

• A deep learning framework utilizing an untrained convolutional neural network (CNN) inspired by deep image priors.
• The method ensures measurement consistency during the phase unwrapping process.
• Experimental validation on both simulated and real complex biological samples.

Main Results:

• The proposed method successfully unwraps the phase for thick and complex samples.
• Accurate phase recovery was demonstrated on both simulated and experimental QPI data.
• The technique provides reliable phase measurements without requiring a training set.

Conclusions:

• The developed deep learning approach offers a robust solution for phase unwrapping in QPI.
• This method significantly advances the capability of QPI for analyzing complex biological specimens.
• It paves the way for more widespread and reliable application of QPI in biological research.