3D Structured Illumination Microscopy via Channel Attention Generative Adversarial Network
Chang Qiao, Xingye Chen, Siwei Zhang, Di Li, Yuting Guo, Qionghai Dai, Dong Li
Abstract
Three-dimensional (3D) structured illumination microscopy (SIM) plays an important role in biological volumetric imaging with the capabilities of doubling the lateral and axial resolution and optical sectioning. However, 3D-SIM suffers from more photobleaching and phototoxicity compared to other volumetric imaging modalities, such as light-sheet microscopy, because it requires 15 raw images per axial slice, which hampers its widespread application in live cell imaging. Here we report the design of a channel attention generative adversarial network (caGAN) that improves the quality of 3D-SIM reconstruction under low signal-to-noise-ratio (SNR) condition and enables reconstruction using fewer raw images. Compared to the conventional algorithm, caGAN-SIM achieves comparable or higher reconstruction fidelity while using 15-fold less signal level. We demonstrate the superior performance of caGAN-SIM for various subcellular structures and its ability in long-term multi-color 3D super-resolution imaging using the example of dynamic interactions between microtubules and lysosomes in live cells.