Intelligent Imaging: Investigating high-performance, interpretable, and self-supervised intelligent bioimage analysis methods. Exploring new architectures that integrate optical imaging with AI to break through classical observational limits and promote new scientific discoveries.

Neural Imaging Robot: Combining imaging with robotics for neural recording in freely behaving animals. This new architecture will become an indispensable technology for revealing the neural mechanism of complex behaviors, and providing technical support for the advancement of brain science and AI.

Quantum Imaging and Quantum-Enhanced Optical Computing: Exploring new paradigms that integrate quantum effects with optical imaging. Designing and building quantum imaging systems to break the standard quantum limit. As an extension, we will empower optical computing with quantum imaging to achieve high-throughput and high-precision photonic AI computing beyond the standard quantum limit.

We have developed a series of advanced intelligent imaging methods to push the boundaries of observational limits, enabling high-speed and high-sensitivity imaging of dynamic biological processes. These methods offer new tools for life science research. Our work has been applied at institutions around the world, including Yale University, Max Planck Institute, and Boston University. It has not only served as a foundational framework that inspires new intelligent imaging strategies, but also provided critical technical support for fundamental scientific discoveries—such as how hypothalamic neurons sense stomach volume through spinal pathways, and how neutrophils utilize TGFβ signaling to defend against external threats.

Our work has been published in high-impact international journals such as Nature Methods, Nature Biotechnology, Nature Computational Science, Light: Science & Applications, etc. Our research has been selected as a highlight paper by MBoC. We were invited to publish a technical commentary in Nature Methods entitled "Challenges and Opportunities in Bioimage Analysis", which was cited in a solo-authored article by Nobel Laureate Eric Betzig.

Xinyang was invited as a guest speaker at CICAI 2023 workshop and has delivered multiple oral presentations at international academic conferences. His research has been featured in major media outlets such as People’s Daily (Overseas Edition) and Guangming Daily. He has received numerous honors and awards, including the Tsinghua Outstanding Doctoral Dissertation Award, Tsinghua Shuimu Scholar, THU/McGovern Award for Outstanding Research Achievement, and the Beijing Outstanding Graduate Award.

1. DeepCAD for Self-supervised Fluorescence Image Denoising: 

We proposed DeepCAD, a spatiotemporal 3D self-supervised denoising method for fluorescence imaging, achieving—for the first time in the world—intelligent noise suppression in neural calcium imaging. This method breaks the shot-noise limit of imaging sensitivity, enabling the full recording of neural dynamics and preventing information loss caused by noise. This work was published in Nature Methods and has been applied to various imaging modalities and biological phenomena. It not only established a foundational framework for neural voltage imaging (Nature Methods, 2023, pp. 1095–1103), but also provided a powerful tool for cutting-edge biological discoveries, including uncovering new mechanisms of gut-brain axis (Cell, 2022, pp. 2478–2494) and novel phenomena in immune signal transduction (Nature, 2025, pp. 740–748).

2. Real-time Fluorescence Image Enhancement:

We developed a real-time denoising method for fluorescence imaging that improves the photon efficiency by tenfold, allowing real-time high-sensitivity observation of neural dynamics and cell migration in various model organisms. For the first time, we resolved the spatiotemporal dynamics of 3D neurotransmitter (ATP) release after brain injury, obtaining firsthand data on the geometric characteristics of release sites. This work was published in Nature Biotechnology and selected as a highlight paper by MBoC.

3. Solving the Challenge of Denoising High-Speed Biological Dynamics: 

We proposed a 3D orthogonal sampling strategy based on spatial redundancy that addresses the dependency of self-supervised denoising on high temporal resolution. We also designed a lightweight transformer network, achieving state-of-the-art performance in fluorescence image denoising. For the first time, we realized noise suppression of 3D calcium imaging, providing a novel tool for dissecting multidimensional and multiscale neural circuit dynamics. This work was published in Nature Computational Science. Prof. Lachlan Whitehead from the University of Melbourne commented it as “provide a promising way forward”.