Clarifying real-time fluorescence microscopy imaging by limiting shot noise using a deep learning method
How much information is lost in the noise? Often randomness – or stochasticity – in weak signals creates a muffling background babble, making it difficult to pick out subtle sights or sounds. Here, researchers use a new approach called DeepCAD-RT to see through the noise when watching immune cells called neutrophils (stained with green fluorescence) moving through an injured mouse brain under a microscope. But they are limited by compromise – signals must be bright enough to capture, but not so bright as to harm the living tissue – the small numbers of light particles or photons often scatter and fall randomly, creating shot-noise. The team fed pairs of consecutive images from the movie on the left into a neural network, training it to spot the tell-tale movements 'real' fluorescent signals. On the right, they cancel the noise – producing a clearer, real-time video of the moving cells. Similar approaches could be applied to other noisy biology.
BPoD stands for Biomedical Picture of the Day. Managed by the MRC London Institute of Medical Sciences the website aims to engage everyone, young and old, in the wonders of biomedicine. Images are kindly provided for inclusion on this website through the generosity of scientists across the globe.
BPoD is also available in Catalan at www.bpod.cat with translations by the University of Valencia.