Phase-augmented deep learning for cell segmentation in wrapped quantitative phase images.

Area of Science:

• Biophysics
• Cell Biology
• Medical Imaging

Background:

• Cell adhesion and detachment are vital for disease diagnosis, treatment, and biomaterials.
• Optical phase imaging offers label-free, continuous cell observation.
• Accurate cell segmentation is critical for quantitative analysis of dynamic cellular events.

Purpose of the Study:

• To develop a robust cell segmentation method for quantitative analysis of dynamic cellular processes using phase imaging.
• To address challenges posed by phase-wrapping artifacts in quantitative phase images.
• To improve the accuracy and efficiency of cell segmentation in label-free microscopy.

Main Methods:

• Developed a phase-augmented deep learning approach utilizing a U-Net architecture.
• Acquired quantitative phase images using modulated optically computer phase microscopy (M-OCPM).
• Implemented a novel data augmentation strategy with global phase shifts to mitigate phase-wrapping artifacts.

Main Results:

• Achieved improved cell segmentation accuracy in wrapped quantitative phase images.
• Successfully distinguished true cell morphology from phase-wrapping artifacts.
• Eliminated the need for image unwrapping, simplifying the segmentation process.
• Enabled quantitative analysis of cell morphology during detachment.

Conclusions:

• The phase-augmented deep learning method provides accurate and efficient cell segmentation for quantitative analysis of dynamic cellular processes.
• This approach overcomes limitations of phase-wrapping artifacts in quantitative phase imaging.
• The developed technique holds significant value for advancing research in cell biology and disease studies.