Image restoration of degraded time-lapse microscopy data mediated by infrared-imaging.

Time-lapse fluorescence microscopy is key to unraveling the processes underpinning biological development and function. However, living systems, by their nature, permit only a limited toolbox for interrogation. Consequently, following time-lapses, expended samples contain untapped information that is typically discarded. Herein we employ convolutional neural networks (CNNs) to augment the live imaging data using this complementary information. In particular, live, deep tissue imaging is limited by the spectral range of live-cell compatible probes/fluorescent proteins. We demonstrate that CNNs may be used to restore deep-tissue contrast in GFP-based time-lapse imaging using paired final-state datasets acquired using infrared dyes and improve information content accordingly. Surprisingly, the networks are remarkably robust over a wide range of developmental times. We employ said network to GFP time-lapse images captured during zebrafish and drosophila embryo/larval development and demonstrate live, deep tissue image contrast.