中国科学院生物物理所李栋、清华大学戴琼海等研究人员合作开发深度学习超分辨显微成像方法。2021年1月21日，《自然—方法学》杂志在线发表了这项成果。

通过多模态结构照明显微镜（SIM），研究人员首先提供了低分辨率（LR）-超高分辨率（SR）图像配对的广泛数据集，并根据结构复杂性、信噪比和放大因子评估了深度学习SR模型。其次，研究人员设计了深度傅里叶通道注意网络（DFCAN），该网络利用不同特征之间的频率含量差异来学习有关各种生物结构高频信息的精确层次表现。第三，研究人员证明了DFCAN的傅立叶域聚焦技术可以在低信噪比条件下可靠地重建SIM卡图像。

研究人员证明，在多色活细胞成像实验中，DFCAN在十倍长的持续时间上可达到与SIM相当的图像质量，这揭示了线粒体嵴和类核苷的详细结构以及细胞器和细胞骨架的相互作用动力学。

据悉，深度神经网络已实现了从LR到SR图像的惊人转换。但是，对于这种深度学习模型是否以及在何种成像条件下都胜过SR显微镜的研究很少。

附：英文原文

Title: Evaluation and development of deep neural networks for image super-resolution in optical microscopy

Author: Chang Qiao, Di Li, Yuting Guo, Chong Liu, Tao Jiang, Qionghai Dai, Dong Li

Issue&Volume: 2021-01-21

Abstract: Deep neural networks have enabled astonishing transformations from low-resolution (LR) to super-resolved images. However, whether, and under what imaging conditions, such deep-learning models outperform super-resolution (SR) microscopy is poorly explored. Here, using multimodality structured illumination microscopy (SIM), we first provide an extensive dataset of LR–SR image pairs and evaluate the deep-learning SR models in terms of structural complexity, signal-to-noise ratio and upscaling factor. Second, we devise the deep Fourier channel attention network (DFCAN), which leverages the frequency content difference across distinct features to learn precise hierarchical representations of high-frequency information about diverse biological structures. Third, we show that DFCAN’s Fourier domain focalization enables robust reconstruction of SIM images under low signal-to-noise ratio conditions. We demonstrate that DFCAN achieves comparable image quality to SIM over a tenfold longer duration in multicolor live-cell imaging experiments, which reveal the detailed structures of mitochondrial cristae and nucleoids and the interaction dynamics of organelles and cytoskeleton.

DOI: 10.1038/s41592-020-01048-5

Source: https://www.nature.com/articles/s41592-020-01048-5