TiSAM: fusing spatial and histopathology foundation models for whole-slide image segmentation

Abstract

Semantic segmentation of whole-slide images (WSIs) quantifies tissue composition and architecture, but existing models require extensive annotation and lose accuracy as the number of tissue classes increases or on heterogeneous, cross-scanner datasets. Here, we present the Tissue Segment Anything Model (TiSAM), which fuses SAM3 spatial features with UNI2-h pathology features within a multi-scale decoder. Because public benchmarks contain few tissue classes, we release MegaMouse, comprising 36,833 annotations of 52 tissue classes from 12 mouse organs, and an 11-class human pancreas dataset. TiSAM achieved the highest overall performance across four public datasets compared to existing convolutional neural network and vision transformer models, maintained high accuracy on MegaMouse, and maintained high accuracy in the human pancreas dataset when trained on as few as 5 annotated WSIs where previous integrations of SAM with pathology foundation models did not. TiSAM segmentations were also more consistent across eight slide scanners compared to other models. These results establish TiSAM as a general and highly adaptable framework for tissue segmentation, enabling studies containing many tissue classes, limited annotations, or images from multiple scanners.

Generalizable virtual staining from histopathology to spatial proteomics across biological scales
Y. Liu, O. Falik, Z. Zhong +9
bioRxiv2 Nov 2026PreprintCC BY-NC-NDFull textBioinformatics & computational biologybioRxiv
Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation
Sally Chen, Roxana Zahedi, Lucy Chhuo +12
arXiv17 Sep 2026PreprintCC BY-SA 4.0Full text1 citedCell biology Genetics & genomicsarXiv
NexuST: A Hierarchical Foundation Model for Spatial Transcriptomics
H. Liu, Q. Zhao, L. Lin +9
bioRxiv22 Sep 2026PreprintCC BYFull textBioinformatics & computational biologybioRxiv
MCseg: AI agent-guided workflow search for no-code cell segmentation and transcript attribution in spatial transcriptomics
C.-R. Chan, N.-W. Chang, C.-Y. Wang +2
bioRxiv20 Sep 2026PreprintCC BY-NC-NDFull textBioinformatics & computational biologybioRxiv
Overcoming segmentation bottlenecks in brain tissue for spatial proteomic inference at population scale
L. You, J. L. Durbin, G. Meng +6
bioRxiv22 Oct 2026PreprintCC BYFull textBioinformatics & computational biologybioRxiv

Cite

BibTeX RIS

BibTeX

@misc{ding2026tisam,
  title = {{TiSAM: fusing spatial and histopathology foundation models for whole-slide image segmentation}},
  author = {Z. Ding and L. Dequiedt and M. Forjaz and V. Matos-Romero and Y. Shen and T. Benducci and A. Forjaz and X. Fei and Y. Zhu and P. Nwoga and W. Dhana and J. Pascalin and S. S. Lee and L. Garza and S. H. Lee and E. J. Pierce and M. I. Canto and J. P. Morton and D. Cihakova and D. Wirtz and Data Science TeamLab and A. L. Kiemen},
  howpublished = {bioRxiv},
  year = {2026},
  month = oct,
  doi = {10.64898/2026.10.16.759688},
  url = {https://www.biorxiv.org/content/10.64898/2026.10.16.759688v1},
  note = {CC BY-NC-ND},
}

RIS

TY  - UNPB
TI  - TiSAM: fusing spatial and histopathology foundation models for whole-slide image segmentation
AU  - Z. Ding
AU  - L. Dequiedt
AU  - M. Forjaz
AU  - V. Matos-Romero
AU  - Y. Shen
AU  - T. Benducci
AU  - A. Forjaz
AU  - X. Fei
AU  - Y. Zhu
AU  - P. Nwoga
AU  - W. Dhana
AU  - J. Pascalin
AU  - S. S. Lee
AU  - L. Garza
AU  - S. H. Lee
AU  - E. J. Pierce
AU  - M. I. Canto
AU  - J. P. Morton
AU  - D. Cihakova
AU  - D. Wirtz
AU  - Data Science TeamLab
AU  - A. L. Kiemen
PY  - 2026
DA  - 2026/10/16
PB  - bioRxiv
DO  - 10.64898/2026.10.16.759688
UR  - https://www.biorxiv.org/content/10.64898/2026.10.16.759688v1
AB  - Semantic segmentation of whole-slide images (WSIs) quantifies tissue composition and architecture, but existing models require extensive annotation and lose accuracy as the number of tissue classes increases or on heterogeneous, cross-scanner datasets. Here, we present the Tissue Segment Anything Model (TiSAM), which fuses SAM3 spatial features with UNI2-h pathology features within a multi-scale decoder. Because public benchmarks contain few tissue classes, we release MegaMouse, comprising 36,833 annotations of 52 tissue classes from 12 mouse organs, and an 11-class human pancreas dataset. TiSAM achieved the highest overall performance across four public datasets compared to existing convolutional neural network and vision transformer models, maintained high accuracy on MegaMouse, and maintained high accuracy in the human pancreas dataset when trained on as few as 5 annotated WSIs where previous integrations of SAM with pathology foundation models did not. TiSAM segmentations were also more consistent across eight slide scanners compared to other models. These results establish TiSAM as a general and highly adaptable framework for tissue segmentation, enabling studies containing many tissue classes, limited annotations, or images from multiple scanners.
ER  -

Abstract

Related

Cite