CenSegNet: a generalist high-throughput deep learning framework for centrosome phenotyping at spatial and single-cell resolution in heterogeneous tissues

Sep 22, 2026

Centrosome abnormalities (CA) are a hallmark of epithelial cancers, yet their spatial complexity and phenotypic heterogeneity are poorly understood due to limits in conventional image analysis. CenSegNet (Centrosome Segmentation Network) provides a modular deep learning framework for high-throughput segmentation of centrosomes and epithelial architecture. Applied to tissue microarrays from 911 breast cancer cores across 127 patients, it enables spatially resolved quantification of numerical and structural CA. CA subtypes show distinct distributions and age-dependent dynamics, associate with tumor grade, hormone receptor status, genomic alterations, and nodal involvement, and link to survival and local aggressiveness.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages24

Reading Time1 h 0 min

AccessFree to download

CenSegNet overview

Purpose and problem scope

Framework design and capabilities

Application to breast cancer tissue microarrays

Biological findings from CA subtypes

Background on centrosomes and CA

Centrosome function in cells

Regulation across the cell cycle

Cancer relevance and clinical associations

Origins of numerical and structural CA

Numerical defects

Structural defects

Mechanistic links and open questions

Existing annotation and analysis approaches

Manual annotation limitations

Semiautomated pipelines and related methods

Article [https://doi.org/10.1038/s41467-026-75393-y](https://doi.org/10.1038/s41467-026-75393-y) CenSegNet: a generalist high-throughput deep learning framework for centrosome phenotyping at spatial and single-cell resolution in heterogeneous tissues Received: 29 October 2025 | Accepted: 30 June 2026 | | --- | | | | Check for updates | Jiaoqi Cheng1,2,10, Keqiang Fan2,3,10, Miles Bailey1,2, Xin Du4, Rajesh Jena5,6, Constantinos Savva7,8, Ewan Reed1,2, Mengyang Gou1,2, Peixin Zuo 9, Ramsey Cutress2,7, Stephen Beers 2,8 , Xiaohao Cai2,3 & Salah Elias 1,2 Centrosome abnormalities (CA) are a hallmark of epithelial cancers, yet their spatial complexity and phenotypic heterogeneity remain poorly understood due to limitations in conventional image analysis. Here we present CenSegNet (Centrosome Segmentation Network), a modular deep learning framework for high-throughput segmentation of centrosomes and epithelial architecture, enabling accurate and generalisable centrosome phenotyping at spatial and single-cell resolution across imaging modalities and tissue contexts. Applied to tissue microarrays comprising 911 breast cancer cores from 127 patients, CenSegNet enables large-scale, spatially resolved quantiﬁcation of numerical and structural CA. We show that these CA subtypes are mechanistically uncoupled, exhibiting distinct spatial distributions, age-dependent dynamics, and associations with tumour grade, hormone receptor status, genomic alterations and nodal involvement. Structural CA are associated with overall survival, whereas discordant CA proﬁles at tumour margins correlate with local tumour aggressiveness and stromal remodelling. These ﬁndings establish CenSegNet as a scalable platform for spatially resolved centrosome phenotyping, enabling systematic investigation of centrosome biology and its dysregulation in cancer and other epithelial diseases. Centrosomes, composed of a pair of orthogonally arranged centrioles surrounded by pericentriolar material (PCM), function as the principal microtubule-organising centres (MTOCs) in animal cells. They play essential roles in diverse cellular processes, including vesicular trafﬁcking, cell polarity, motility, ciliogenesis, and the assembly of a bipolar mitotic spindle during cell division1,2. Centrosome number and size are tightly regulated during the cell cycle, with duplication occurring once per cell cycle during S phase, ensuring the formation of the mitotic spindle and equal inheritance of chromosomes by daughter cells3–5. 1School of Biological Sciences, University of Southampton, Southampton, UK. 2Institute for Life Sciences, University of Southampton, Southampton, UK. 3School of Electronics and Computer Science, University of Southampton, Southampton, UK. 4Cavendish Laboratory, Department of Physics, University of Cambridge, Cambridge, UK. 5Department of Oncology, University of Cambridge, Cambridge, UK. 6Cambridge University Hospitals NHS Foundation Trust, Cambridge, UK. 7University Hospital Southampton NHS Foundation Trust, Southampton, UK. 8Centre for Cancer Immunology, School of Cancer Sciences, University of Southampton, Southampton, UK. 9School of Mathematical Sciences, University of Southampton, Southampton, UK. 10These authors contributed equally: Jiaoqi Cheng, Keqiang Fan. e-mail: [s.a.beers@soton.ac.uk](s.a.beers@soton.ac.uk); [x.cai@soton.ac.uk](x.cai@soton.ac.uk); [s.k.elias@soton.ac.uk](s.k.elias@soton.ac.uk) Centrosome abnormalities (CA) can lead to multipolar spindle formation, chromosomal missegregation, and aneuploidy3,6,7—a hallmark of cancer3,4,7. The hypothesis that CA-induced aneuploidy contributes to tumorigenesis was ﬁrst proposed by Theodor Boveri over a century ago8. In recent years, CA have been documented in several solid tumours including breast, prostate, colon, ovarian, and pancreatic cancers3,7,9–13, as well as haematological malignancies such as multiple myeloma, lymphomas, and leukaemias14,15. While their role in tumour initi

FAQs

What is CenSegNet and what problem does it address?

CenSegNet is a modular deep learning framework for high-throughput segmentation of centrosomes and epithelial architecture. It addresses the lack of robust tools that can resolve centrosome phenotypes within complex tissue spatial context and at single-cell resolution.

How was CenSegNet evaluated in this study?

The framework was applied to tissue microarrays containing 911 breast cancer cores from 127 patients. This enabled large-scale, spatially resolved quantification of numerical and structural centrosome abnormalities.

What do the CA subtypes reveal about cancer biology?

The study reports that CA subtypes are mechanistically uncoupled, with distinct spatial distributions and age-dependent dynamics. Numerical and structural CA are associated with tumor grade, hormone receptor status, genomic alterations, nodal involvement, and differences in survival and local aggressiveness.

Sep 22, 2026

Upload by RuangKosong