Generative super-resolution AI accelerates nanoscale analysis of cells

Super-resolution microscopy (SRM) surpasses Abbe's diffraction limit, thus enabling nanoscale observation of cells. However, SRM techniques, such as stochastic optical reconstruction microscopy (STORM), suffer from long acquisition times which can significantly impact imaging throughput. To address this issue, we adapted the enhanced super-resolution generative adversarial network from natural to microscopy images. Our goal is to generate super-resolution images from widefield microscopy images in shorter times. We implemented this for imaging microtubules of cells to obtain STORM-like images. Different models were trained by using transfer learning and progressive fine-tuning. The generated images, evaluated by peak signal-to-noise ratio, structural similarity index and expert human validation, prove that this deep learning approach is suitable for microscopy, allowing for 4x-higher throughput of nanoscale imaging compared to unsupported techniques.