PostFocus: automated selective post-acquisition high-throughput focus restoration using diffusion model for...

Area of Science:

• Cell biology
• Microscopy
• Computational imaging

Background:

• High-throughput time-lapse imaging is crucial for single-cell analysis.
• Label-free phase-contrast microscopy offers noninvasive live-cell imaging.
• Out-of-focus cells are common in time-lapse sequences due to cell migration.

Purpose of the Study:

• To develop a computational method for correcting out-of-focus images in live-cell microscopy.
• To improve the accuracy of cell analysis from time-lapse video data.

Main Methods:

• Implemented a classifier to detect out-of-focus images.
• Utilized a de-noising diffusion probabilistic model to restore in-focus images.
• Applied the method to time-lapse sequences of migrating immune cells.

Main Results:

• The diffusion model outperformed deep discriminative models in image quality and cell boundary definition.
• PostFocus enhanced the accuracy of cell detection and contact identification.
• The method increased the yield of usable videos from time-lapse experiments.

Conclusions:

• PostFocus effectively corrects out-of-focus images in label-free live-cell microscopy.
• The approach improves downstream image analysis accuracy and data yield.
• This method enhances the utility of high-throughput time-lapse imaging for cell biology research.