RESOLFT time lapse imaging empowered by deep learning

Capturing the dynamic interplay of proteins and organelles within cells requires microscopy methods with high spatial and temporal resolution. RESOLFT nanoscopy demonstrated great potential for live cell recording by using patterned light and reversibly switching fluorescent probes. Despite the latest improvement in both switching probes and optical schemes, the length and speed of the time-lapse is still limited by photobleaching and sampling. Here, we present two new approaches based on deep learning to restore low-SNR or under-sampled images, which pushes the experimental boundaries of RESOLFT nanoscopy achieving up to 5 times longer imaging with 10 times lower dose of light per frame, or a 4-fold increase in imaging speed. The increased speed and length of time lapses enable to follow vesicles budding in and out from intact mitochondria, as well the actin meshwork formation and dissociations opening up new possibility for dynamic data acquisition at the sub-organelle level.