analiza aiqbench
This paper addresses a critical bottleneck in medical AI by using spatial transcriptomics data as weak supervision to train nuclei classifiers, significantly reducing reliance on expensive manual annotations. The approach demonstrates that transferring cell-type identities from gene expression profiles can achieve performance comparable to fully supervised methods across unseen tissues. This advancement offers a scalable pathway for deploying deep learning models in histopathology where curated datasets are scarce.
Leveraging Cell-level Spatial Transcriptomics as Weak Supervision for Nuclei Classification
arXiv:2604.23481v2 Announce Type: replace-cross Abstract: Deep learning-based nuclei analysis in histopathology relies heavily on large-scale, expert-curated datasets that are costly and difficult to scale across organs. To address this, we propose a scalable framework leveraging cell-level spatial transcriptomics (ST) as an alternative supervision source for nuclei classification. Cell-level ST provides both gene expression profiles and nucleus masks, enabling automatic generation of image-label pairs without manual annotation. We convert gene expression into cell-type labels and introduce an image-oriented classification strategy to effectively transfer these identities to histopathology images. Evaluated on unseen tissues, our framework achieves comparable segmentation and classification performance compared with conventional supervised approaches, despite using fewer manually annotated datasets.
Abstract:Deep learning-based nuclei analysis in histopathology relies heavily on large-scale, expert-curated datasets that are costly and difficult to scale across organs. To address this, we propose a scalable framework leveraging cell-level spatial transcriptomics (ST) as an alternative supervision source for nuclei classification. Cell-level ST provides both gene expression profiles and nucleus masks, enabling automatic generation of image-label pairs without manual annotation. We convert gene expression into cell-type labels and introduce an image-oriented classification strategy to effectively transfer these identities to histopathology images. Evaluated on unseen tissues, our framework achieves comparable segmentation and classification performance compared with conventional supervised approaches, despite using fewer manually annotated datasets.
Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2604.23481 [cs.CV]
(or arXiv:2604.23481v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2604.23481 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 203 cuvinte
Subiecte conexe
LATTICE: Graph Self-Supervised Learning for Multimodal Spatial Omics Integration
arXiv · 21 iul. 2026
SPATIA: Multimodal Generation and Prediction of Spatial Cell Phenotypes
arXiv · 17 iun. 2026
LASA: A Weak Supervision Method for Open-Vocabulary Scene Sketch Semantic Segmentation
arXiv · 11 iun. 2026
SceneConductor: 3D Scene Generation from a Single Image with Multi-Agent Orchestration
arXiv · 17 iun. 2026