Revealing architectural order with quantitative label-free imaging and deep learning

Area of Science:

• Biomedical Imaging
• Cell Biology
• Neuroscience

Background:

• Label-free imaging techniques are crucial for studying live biological systems without exogenous labels.
• Existing methods often lack the resolution or multiplexing capabilities to capture complex cellular and tissue architectures.
• Understanding tissue microstructure, such as axon orientation, is vital in neuroscience and developmental biology.

Purpose of the Study:

• To develop a quantitative label-free imaging method for simultaneous measurement of density, anisotropy, and orientation.
• To integrate this method with deep neural networks for predicting fluorescence images.
• To demonstrate its utility in visualizing structures not visible with conventional imaging and rescuing experimental artifacts.

Main Methods:

• Quantitative label-free imaging with phase and polarization (QLIPP) was employed.
• Deep neural networks, specifically a multi-channel 2.5D U-Net architecture, were utilized for image prediction.
• Novel data normalization techniques were developed for accurate prediction of myelin distribution.

Main Results:

• QLIPP successfully measured density, anisotropy, and orientation in unlabeled live cells and tissue.
• Predicted fluorescence images revealed anatomical regions and axon tract orientation in human brain tissue, surpassing brightfield imaging.
• The method demonstrated the ability to rescue experimental defects in tissue labeling.
• Computational efficiency was achieved for 3D predictions over large fields of view.

Conclusions:

• QLIPP offers a powerful label-free approach for high-resolution imaging of cellular and tissue architecture.
• The integration with deep learning enhances predictive capabilities and broadens imaging applications.
• This technique holds promise for advancing studies of structural organization from the organelle to tissue level.