Source of information: Prof. Chen Liangyi's Team

In recent years, the development of super-resolution fluorescence microscopy has greatly advanced the study of subcellular structures. Among them, structured illumination microscopy (SR-SIM) based on structured light illumination is particularly suitable for super-resolution imaging of live cells due to its high photon conversion efficiency. However, in 2018, the work of Chen Liangyi's laboratory pointed out that although structured light has higher photon conversion efficiency for super-resolution than other types of microscopy, the deconvolution reconstruction process it involves amplifies noise and generates artifacts, thus affecting the credibility and quantitative analysis of super-resolution images. After proposing an iterative reconstruction method based on Heisenberg regularization term utilizing spatiotemporal continuity as prior knowledge, many other research groups have also developed different methods to suppress artifacts. Existing methods based on physical models or general prior knowledge can suppress artifacts caused by noise, but they cannot completely suppress artifacts caused by background defocus or light scattering. On the other hand, although deep neural network reconstruction methods can better suppress various artifacts caused by reconstructed super-resolution images, they may introduce local distortions and lower the resolution.

In January 2023, Chen Liangyi's team at Peking University combined the advantages of physical models and deep learning reconstruction methods. They proposed a hybrid reconstruction method (TDV-SIM) by utilizing Total Deep Variation (TDV) network as the regularization term of the reconstruction objective function and combining it with SIM physical models. This method can suppress artifacts while maintaining resolution. When processing images of different cell structures, TDV-SIM can better preserve the true signal compared to pure deep learning methods, and it can more effectively remove artifacts compared to methods based on physical models. The related work titled "Hybrid reconstruction of the physical model with the deep learning that improves structured illumination microscopy" (doi: 10.1117/1.APN.2.1.016012) was published in Advanced Photonics Nexus.

Wang Jianyong, a master's graduate from the School of Software and Microelectronics at Peking University, Fan Junchao, an associate professor from the School of Computer Science at Chongqing University of Posts and Telecommunications, and Zhou Bo, a doctoral student from the Future Technology Institute at Peking University, are the co-first authors of this paper. Professor Chen Liangyi from the Future Technology Institute at Peking University and Researcher Huang Xiaoshuai from the Biomedical Engineering Department at Peking University served as the corresponding authors of this paper. This research was supported by the National Key Science and Technology Special Projects, the National Natural Science Foundation of China, the Beijing Natural Science Foundation, and the Lingang Laboratory, among others.