Single-Molecule-Localization-Microscopy (SMLM) techniques like dSTORM and PALM are state-of-the-art tools for biologists to gain quantitative and qualitative information of cells beyond the Abbe diffraction-limit. Typically, these methods are demanding in terms of photon sensitive cameras, typically found in upscale sCMOS or emCCD chips.
But what if one could use off-the-shelf components to realize super-resolution on a budget? The development of mobile phones created surprisingly powerful camera-sensors and computational resources worth considering as cost-effective imaging and processing devices for wide variety of microscopy techniques.
Furthermore, recent advances in the field of machine-lear ning enables e.g. localizing the fluorophore’s position using neural networks. We adapted this idea and build a generative adversarial network (GAN) which serves as a robust localization-method. It successfully differentiates between true blinking events and compression-/noise-artifacts e.g. coming from a cellphone’s video-stream. The network-approach makes it further possible to split and distribute the steps of training and inference. Thus, it is possible to directly localize the blinking events acquired with a commercially available cellphone on the same device to produce a localization map with an optical resolution well below 80nm.
Our method can be applied to any research microscope equipped with the appropriate laser-illumination. This does not only reduce the system’s complexity, but dramatically reduces the price. This is especially useful in educational environments where expensive camera is rarely available. We believe, that cellSTORM paves the way to involve a bigger mass of people and thus democratizing science.
contact: Benedict Diederich