Deep-Z

Deep-Z reconstructs three-dimensional fluorescence volumes from single two-dimensional wide-field fluorescence microscopy images by using a trained deep neural network to virtually refocus images and apply time-reversal of fluorescence propagation.

Key Features:

• Virtual Refocusing: A trained deep neural network propagates a single 2D fluorescence image across specified 3D planes to produce a volumetric representation without mechanical axial scanning.
• Enhanced Depth-of-Field: Increases depth-of-field by up to 20-fold, enabling comprehensive 3D imaging from images captured at a single focal plane.
• Correction Capabilities: Digitally corrects for sample drift, tilt, and optical aberrations post-acquisition to improve reconstruction fidelity.
• Cross-Modality Integration: Enables 3D refocusing of wide-field fluorescence images to align with confocal microscopy data acquired at different sample planes.

Scientific Applications:

• Neuroscience imaging: Applied to imaging neuronal activity in Caenorhabditis elegans to provide 3D visualization of dynamic biological processes and complex structures in living organisms.

Methodology:

The method trains a deep neural network on fluorescence wave propagation and time-reversal to digitally project a single 2D image onto multiple 3D planes, eliminating the need for traditional axial scanning.