Fast light-field 3D microscopy with out-of-distribution detection and adaptation through Conditional Normalizing...

Area of Science:

• Neuroscience
• Biophysics
• Microscopy

Background:

• Real-time 3D fluorescence microscopy is vital for analyzing live organisms, particularly neural activity.
• The eXtended field-of-view light field microscope (XLFM) offers single-snapshot 3D imaging but suffers from slow traditional reconstruction.
• Existing neural network methods are fast but lack crucial certainty metrics for biomedical applications.

Approach:

• A novel conditional normalizing flow architecture is proposed for rapid 3D reconstruction of XLFM data.
• This method achieves 8 Hz reconstruction of 512 × 512 × 96 voxel volumes.
• Training requires minimal data (10 image-volume pairs) and takes under two hours.

Key Points:

• The approach enables fast, high-resolution 3D neural activity reconstruction.
• Normalizing flows provide exact likelihood computation for distribution monitoring and out-of-distribution detection.
• The method ensures trustworthiness through certainty metrics, unlike other fast neural network approaches.

Conclusions:

• The proposed conditional normalizing flow significantly accelerates 3D reconstruction for XLFM.
• This technique enhances the reliability and speed of live neural activity monitoring.
• It offers a trustworthy solution for real-time spatiotemporal analysis in the biomedical field.