Junchao Zhu; Ruining Deng; Junlin Guo; Tianyuan Yao; Siqi Lu; Chongyu Qu; Juming Xiong; Yanfan Zhu; Zhengyi Lu; Yuechen Yang; Marilyn Lionts; Yucheng Tang; Daguang Xu; Yu Wang; Shilin Zhao; Haichun Yang; Yuankai Huo. (2026).听.听Briefings in Bioinformatics, 27(3), bbag255.听
Spatial transcriptomics is a technology that measures which genes are active in a tissue while also showing where those gene signals come from inside the tissue. This gives scientists a much clearer picture of how tissues are organized and how different parts work together. So far, much of the progress in this field has come from bioinformatics, but many methods still treat space mainly as a set of coordinates and relationships, without fully using the detailed visual information available in tissue images. This paper reviews a newer approach that uses computer vision, a branch of artificial intelligence that analyzes images, to make better use of that visual detail and connect tissue shape and structure with gene activity. These methods could help address some major limits of spatial transcriptomics, such as its high cost, limited use in clinical settings, and the fact that many analyses are still based on two-dimensional images even though real tissues are three-dimensional. For example, some models can estimate spatial gene patterns directly from microscope images, which could reduce the need for expensive sequencing while still capturing important biological information. Other methods can rebuild three-dimensional tissue maps from image data. The paper is the first broad review of computer vision methods for spatial transcriptomics, organizing them by model type, learning approach, task, and dataset, and highlighting the main challenges and future opportunities in this fast-growing area.
Figure 1
Overview of some well-known vision-driven models for ST from 2020 to 2025.