Self-supervised learning for multiplexing super-resolution confocal microscopy

Abstract

Confocal microscopy has long been a cornerstone technique for visualizing complex interactions and processes within cellular structures. However, achieving super-resolution imaging of multiple organelles and their interactions simultaneously has remained a significant challenge. Here, we present a self-supervised learning approach to transform diffraction-limited, single-colour input images into multi-colour super-resolution outputs. Our approach eliminates the need for paired training data by utilizing a degradation model. By enhancing the resolution of confocal images and improving the identification and separation of cellular targets, this method bypasses the necessity for multi-wavelength excitation or parallel detection systems. Trained on an extensive dataset, the model effectively distinguishes and resolves multiple organelles with high fidelity, overcoming traditional imaging limitations. This technique requires no hardware modifications, making multi-colour super-resolution imaging accessible to any standard confocal microscope. We validated its performance by demonstrating two- and three-colour super-resolution imaging of both fixed and live cells. The technique offers a streamlined and data-efficient solution for multi-channel super-resolution microscopy, while opening new possibilities for investigating dynamic cellular processes with unprecedented clarity.