Spatially resolved transcriptomics (SRT) can measure gene activity inside intact tissue. But a key step — deciding which measured RNA belongs to which cell — is still unresolved. In a community paper, the authors argue that cell segmentation should not be treated as a routine preprocessing step. They say it is a central and unsolved problem that adds major uncertainty to SRT studies.

The paper reviews why segmentation is hard for SRT. Unlike standard microscopy, SRT combines images with sparse, count-based RNA measurements. RNA spots can be shifted by experimental effects such as transcript diffusion or errors in locating molecules. Tissue slicing can cut cells, producing non‑cellular RNA. Many tissues lack clear membrane markers. And three‑dimensional tissue structure is often squeezed into two‑dimensional images. These issues create mixed or noisy cell profiles that are hard to assign reliably.

The authors summarize current methods and their limits. Modern SRT platforms such as Xenium, MERSCOPE, CosMx, Stereo‑seq, Visium HD, and Open‑ST can reach cellular or subcellular resolution. There are more than a dozen segmentation-related tools, and some recent approaches try to assign transcripts and cell boundaries at the same time — called transcript‑informed segmentation. But it is unclear which method works best across different technologies and tissues. Improved stains and AI image methods have helped, yet the core challenges remain.

Why this matters is clear: wrong segmentation changes the measured gene profile of cells. Those errors can bias cell‑type labels, distort inferred cell‑cell interactions, and mislead biological conclusions. The authors stress that segmentation mistakes can propagate through every downstream analysis that depends on cell‑level data.