1082-Hist2Cell: Cellular Architectures from Histology Images

Hist2Cell is an innovative deep learning framework designed to predict detailed cellular architectures and fine-grained cell types directly from standard H&E histology images. By utilizing a graph-transformer architecture, the model captures both local tissue features and long-range spatial correlations to map up to 80 distinct cell types without requiring expensive molecular data for every sample. Research shows that this technology achieves high accuracy in identifying cell-type abundance and spatial colocalization across diverse tissues, including lung, breast, and skin. Beyond mapping, Hist2Cell enhances cancer prognosis by providing survival risk predictions that are comparable to, or better than, traditional bulk RNA-sequencing. The framework also enables the creation of super-resolved cellular maps, offering a cost-effective tool for large-scale clinical diagnostics and personalized medicine. Ultimately, this approach bridges the gap between routine pathology and complex spatial transcriptomics.

References:

• Zhao W, Liang Z, Huang X, et al. Hist2cell: deciphering fine-grained cellular architectures from histology images[J]. Cell Genomics, 2026, 6(3).