EUVIMEDEUVIMED entdecken
ÜbersichtsarbeitOpen Access

Transforming subcellular spatial transcriptomics: deep learning models for cell segmentation

Isabelle A Rathbun, Nirad Banskota, Elin Lehrmann iD, Myriam Gorospe iD, Supriyo De iD

Nucleic Acids Research · 10.08.2026 · 54 (15)

DOI: 10.1093/nar/gkag782 ↗

Kurz gefasst

Übersichtsarbeit in der Zeitschrift Nucleic Acids Research, online erschienen am 20.08.2026. Verfasst von Isabelle A Rathbun und 4 weiteren Autor:innen. Beteiligte Einrichtung: Laboratory of Genetics and Genomics (LGG), National Institute on Aging Intramural Research Program (NIA IRP), National Institutes of Health (NIH), Ba…. Hauptthemen laut MEDLINE: Deep Learning, Spatial Transcriptomics. Die Arbeit nennt 69 Literaturangaben. Der Volltext ist frei in PubMed Central verfügbar. Gefördert von: NIH HHS, IRP, National Institutes of Health.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Subcellular spatial transcriptomics (SSTs) is transforming biology by revealing where individual RNA molecules are located within intact tissues, providing unprecedented insight into how cells function and interact. Achieving this promise depends on accurately identifying the boundaries of individual cells, making cell segmentation one of the field’s central computational challenges. A core methodological issue is how best to ensure correct and biologically accurate assignment of transcripts to cells using cell segmentation algorithms. Recent advances in deep learning models have been instrumental in developing tools to enable more accurate transcript-to-cell maapping across a wider range of tissues and resolutions. Here, we review key features of emerging deep learning-based cell segmentation strategies, including fully convolutional neural networks, transformer-based models, and foundation models, highlighting their architectural innovations, performance characteristics, and practical limitations. We conclude that transformer-based and foundation model-based models have better generalizability and require minimal retraining, but their adoption is limited by larger data requirements and higher computational cost. We predict a larger adoption of these more flexible and robust methods as deep learning architectures become more scalable and improved hardware accessibility permits further advancements of quantitative, high-resolution tissue analysis. We propose that the convergence of biology and artificial intelligence (AI) and the continued innovation in deep learning-based cell segmentation will accelerate method development, standardization, and deployment in both basic biological research and clinical applications.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Deep LearningSpatial TranscriptomicsConvolutional Neural NetworksAnimalsHumansAlgorithms

Autor:innen und Institutionen

  • Laboratory of Genetics and Genomics (LGG), National Institute on Aging Intramural Research Program (NIA IRP), National Institutes of Health (NIH), Baltimore, MD 21224, United States.
  • Department of Computer Science, University of Maryland, College Park, MD 20742,United States.

ORCID-Profile: Elin Lehrmann, Myriam Gorospe, Supriyo De

Förderung

  • NIH HHS
  • IRP
  • National Institutes of Health

Zitieren

APA

Rathbun, I. A., Banskota, N., Lehrmann, E., Gorospe, M., & De, S. (2026). Transforming subcellular spatial transcriptomics: deep learning models for cell segmentation. Nucleic Acids Research, 54(15). https://doi.org/10.1093/nar/gkag782

Vancouver

Rathbun IA, Banskota N, Lehrmann E, Gorospe M, De S. Transforming subcellular spatial transcriptomics: deep learning models for cell segmentation. Nucleic Acids Research. 2026;54(15). doi:10.1093/nar/gkag782

BibTeX
@article{Rathbun2026,
  author = {Rathbun, Isabelle A and Banskota, Nirad and Lehrmann, Elin and Gorospe, Myriam and De, Supriyo},
  title = {Transforming subcellular spatial transcriptomics: deep learning models for cell segmentation},
  journal = {Nucleic Acids Research},
  year = {2026},
  volume = {54},
  number = {15},
  doi = {10.1093/nar/gkag782}
}
Kostenlos starten →