Fast, live-cell imaging of 15 intracellular compartments by deep learning segmentation of super-resolution data

Abstract

The number of colors that can be used in fluorescence microscopy to image the live-cell anatomy and organelles interactions is far less than the number of intracellular organelles and compartments. Here, we report that deep convolutional neuronal networks can predict 15 subcellular structures from super-resolution spinning-disk microscopy images using only one dye, one laser excitation, and two detection channels. Comparing to the colocalization images, this method achieves pixel accuracies of over 91.7%, which not only bypasses the fundamental limitation of multi-color imaging but also accelerates the imaging speed by more than one order of magnitude.