Instant two-image diffractive lattice structured illumination microscopy using transfer learning

Conventional structured illumination microscopy (SIM) requires at least nine raw images to reconstruct a super- resolution image, which limits its temporal resolution and applicability to dynamic samples. To overcome this limitation, we have developed a two-image diffractive lattice SIM (TIDL-SIM) using transfer learning. It uses only two images, a wide-field image and a diffractive lattice illumination modulated image, to construct the super- resolution image. A deep neural network extracts low spatial frequency information from the wide-field image and high spatial frequency information from the modulated image. It is trained on an auxiliary domain consisting of simulated images, but is transferable to the target task of reconstructing experimental SIM images. Finally, the TIDL-SIM achieves super-resolution live-cell imaging with acquisition time of 40 milliseconds and construction time of approximately 29 milliseconds per frame. The reduced acquisition and construction time will facilitate the advancement of real-time SIM.