.png){kind=icon}
.png){kind=icon}
Return
High-speed blind structured illumination microscopy via unsupervised algorithm unrolling
.png){kind=icon}
OA
AI
DOI：10.1038/s41467-026-68693-w.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Blind structured illumination microscopy (blind-SIM) is a valuable tool for achieving super-resolution without the need for known illumination patterns. However, in its current formulation the algorithm requires many iterations to converge, leading to long inference times and limited use for real-time or video-rate imaging. We present unrolled blind-SIM (UBSIM), an algorithm which integrates a learnable neural network inside the unrolled iterations of the blind-SIM algorithm. UBSIM delivers a reconstruction speed two to three orders of magnitude faster than that of current iterative blind-SIM methods, while achieving similar resolution and image quality. Furthermore, we demonstrate that UBSIM can be trained in an unsupervised manner that reduces hallucinations and produces superior generalization capability when compared to benchmark super-resolution networks. We test UBSIM experimentally on live cells and present video-rate super-resolution imaging up to 50 Hz. Using our method, we observe dynamic remodeling of the endoplasmic reticulum with high spatiotemporal resolution. Burns et al. introduce an unrolled, physics-informed machine learning method that speeds up blind structured illumination microscopy by orders of magnitude while preserving generalizability, enabling real-time superresolution imaging in live cells.
Keywords:
blind-SIM
super-resolution
unrolled algorithm
structured illumination microscopy
unsupervised learning
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
15.7
Papers:
9.3W
Citations:
91.2W